docs: correct claims in CLAUDE.md, CHOOM-INIT.md and the etc/ deployment READMEs (#658, #659)

[vibechoom]

Docs-only drift correction sweep closing #658 and #659. Every claim in both issues was re-verified against the current tree (my own read of the doc sentence + the code it describes) before editing — none were taken on the issue's word alone.

CLAUDE.md

• Before: :197 overlays list = "(frame, notifications, osd, prompt, sidebar)" — 5 modules.
  After: adds consent.
  Evidence: trollshell/src/overlays/mod.rs:8-13 declares 6 pub mods: consent, frame, notifications, osd, prompt, sidebar.
• Before: :203 names plugins.rs as the host-transport file.
  After: plugins/ (module dir since chore(shell): structure cleanups — dead lock-screen CSS, audio row duplication, visibility-gate triplication, plugins.rs split #443, not a single file).
  Evidence: trollshell/src/plugins.rs doesn't exist; ls trollshell/src/plugins/ shows mod.rs + 8 submodules. Commit 6ddf7d9 ("split plugins.rs into a module dir (chore(shell): structure cleanups — dead lock-screen CSS, audio row duplication, visibility-gate triplication, plugins.rs split #443)…") is the split; issue chore(shell): structure cleanups — dead lock-screen CSS, audio row duplication, visibility-gate triplication, plugins.rs split #443's own title is literally "…plugins.rs split" (closed).

docs/CHOOM-INIT.md

• Before: :35-38 "CI does NOT run clippy… local clippy is the only clippy enforcement."
  After: CI DOES run it — rewrote to the accurate nix flake check roster (clippy w/ -D warnings, treefmt, system-tests bucket, both module-eval checks, both nixosTests, options-doc, the two package builds) and pointed at CLAUDE.md's own CI section instead of re-enumerating.
  Evidence: flake.nix:287-294, checks.clippy = craneLib.cargoClippy (... cargoClippyExtraArgs = "--all-targets --features system-tests -- -D warnings"). Added in 06c47b0 ("ci: add cargoClippy check to flake checks", closes CI doesn't run clippy — pedantic-lint regressions merge silently #74) — nine days before ce19c67 created this doc.
• Before: :163-164 cites etc/niri/blur.kdl as live, "hand-merged into Annika's nixos config.kdl."
  After: kept the still-true blur-geometry lesson, replaced the dead citation with a note that the whole frosted-glass experiment (incl. blur.kdl and the hytte-blur crate) was retired.
  Evidence: ls etc/niri/ = binds.kdl frame.kdl README.md session.kdl (no blur.kdl); grep -rn 'blur|xray' etc/niri/ empty. Deleted in 86f5720 (closing sidebar leaves behind blured region. #312, "finish the frost retirement").
• Before: :184-186 lists "PAM file" among what etc/ ships.
  After: clarifies the PAM file is NOT shipped, it's a manual step, citing etc/README.md:57.
  Evidence: find etc -iname '*pam*' → empty. etc/README.md:57-59 and etc/systemd/user/README.md:359-361 both already instruct the deployer to create /etc/pam.d/swaylock by hand.

etc/systemd/user/README.md

• Before: :202-204 "A denied auth/data request raises one informational toast… interactive Allow/Deny prompting is a later phase."
  After: the first request from an agent with no standing grant fires an interactive consent prompt; only a settled standing deny answers silently with a toast.
  Evidence: crates/hytte-plugin-infobroker/src/broker.rs module doc: "Consent policy (phase 1b — interactive Allow/Deny prompting…)"; AuthOutcome::NeedsConsent (~line 202), BrokerState::apply_consent (line 678); trollshell/src/overlays/consent.rs is the prompt overlay (exists).
• Before: :217-220 "The marquee currently shows a decorative banner… now-playing metadata isn't reachable out-of-process yet."
  After: the marquee scrolls the live track off StateKey::NowPlaying, falling back to the banner only when nothing plays.
  Evidence: crates/hytte-plugin-proto/src/manifest.rs:99 StateKey::NowPlaying, :153 Capability::NowPlaying; crates/hytte-plugin-proto/src/state.rs:103 struct NowPlaying; crates/hytte-plugin-audio-widget/src/main.rs's own module doc credits this to feat(plugin-proto): a now-playing StateKey so plugins can show the current track (#506 follow-up) #528 and documents the banner as the fallback (not the current behavior). Both StateKey::NowPlaying and #528 are cited directly in that source file's own comments, not inferred — cross-checked against issue feat(plugin-proto): a now-playing StateKey so plugins can show the current track (#506 follow-up) #528's title ("a now-playing StateKey so plugins can show the current track", closed) and manifest.rs:92's own (#528) citation.

etc/calendar/README.md

• Before: "No recurring-event expansion past the master entry… needs a new hytte-ecal binding — a follow-up, not yet wired."
  After: bullet deleted outright (shipped, nothing to reword).
  Evidence: crates/hytte-ecal/src/lib.rs:459 pub fn generate_instances(...), called from crates/hytte-services/src/calendar.rs:333; exercised by crates/hytte-ecal/examples/probe.rs; the eds-nixos-test nixosTest in flake.nix seeds a FREQ=DAILY;COUNT=5 fixture, so nix flake check covers this path on every run.

etc/kanshi/README.md

• Before: "No per-output wallpaper. The shipped wallpaper service paints a single image across every output… per-output backgrounds are a follow-up."
  After: bullet deleted outright (shipped).
  Evidence: crates/hytte-services/src/wallpaper.rs:223 pub outputs: BTreeMap<String, String>, :278 paints per output from it.
  Note on the sibling doc: docs(etc): four deployment READMEs describe shipped features as unshipped, and #219's GAction verbs are bound nowhere #659 flagged this as contradicting etc/wallpaper/README.md. I read that file — it already correctly documents the outputs JSON key (:27-38) and even calls out "per-output image selection is supported" (:138). No false claim there, so no edit was needed; the contradiction was one-sided (kanshi's false negative) and is resolved by this deletion alone.

etc/niri/README.md — the #219 split (per instructions, NOT unilaterally deciding the config half)

• Before: :122-128 cites page_power_menu in trollshell/src/widgets/pages.rs (doesn't exist) and says reaching the power menu from a keybind "would need a tiny trollshell IPC channel that doesn't exist yet."
  After: corrected the path to panel_power_menu in trollshell/src/panels/power_menu.rs, and replaced the false IPC claim with the real verbs (open-page, power-menu, toggle-sidebar, toggle-recording, all registered as org.gtk.Actions in trollshell/src/commands.rs) plus the exact busctl invocation lifted verbatim from that file's own module doc.
  Evidence: trollshell/src/panels/power_menu.rs:28 pub fn panel_power_menu(); trollshell/src/commands.rs:15-24,57-92 (the module doc's busctl example + the four ActionEntry registrations). git log -- etc/niri/binds.kdl confirms feat(shell): d-bus command surface so niri keybinds can drive sidebar, drawer pages, and power menu #219's commit (93f7961) never touched it — its own commit message says "Wiring the busctl calls into etc/niri keybinds is a follow-up."
  Decision, explicit per instructions: I did not add a keybind to etc/niri/binds.kdl. That file mirrors Annika's personal chords, and which key (if any) maps to which verb is her preference, not a defect — so I documented that the verbs exist and are bindable, gave the working invocation, and left the actual binding to her. binds.kdl is untouched.

Claims that did not hold

None. Every claim cited in #658 and #659 was re-verified against the current tree (doc sentence + the code/file it describes) and confirmed true before I acted on it.

Gates

This diff is markdown-only (git show --stat on the commit: CLAUDE.md, docs/CHOOM-INIT.md, etc/calendar/README.md, etc/kanshi/README.md, etc/niri/README.md, etc/systemd/user/README.md — no .rs/.nix files). What I actually ran:

• nix fmt on the six changed files — applied cleanly (one cosmetic reflow inside a long inline-code span in CHOOM-INIT.md; treefmt/prettier correctly declines to indent continuation lines inside a backtick span, since indentation there would become literal text).
• Local nix build .#trollshell was deliberately skipped (started, then killed once I confirmed it couldn't tell us anything): nix/package.nix:46-50's crane source filter is craneLib.filterCargoSources path type || lib.hasSuffix "assets/hytte-ui/style.css" path — it keeps Rust sources, Cargo.toml/Cargo.lock, and exactly one CSS file. Markdown is not in that filter, so CLAUDE.md/docs//etc/ never reach the derivation and a markdown-only diff produces a bit-identical trollshell-workspace drv. Running it would only burn a from-scratch workspace compile (the remote builder was down, forcing local) for zero signal — this is the same crane-filter trap noted in prior sweeps (assets/ include_str! changes hit the identical gap).
• Clippy/test gates were skipped for the same reason — nothing in this diff is visible to cargo or crane.
• CI's flake-check (which does build the real derivations) is the actual gate for this PR and will run on push.

I touched neither docs/live-verify.md (out of scope, sequenced after #644/#645 per instructions) nor any .rs file (Rust lanes are owned by open PRs #644/#645 and a live agent on #643) nor etc/niri/binds.kdl (see the #219 decision above).

Closes #658
Closes #659