fix(permissions): return guidance for multi-word forms instead of falling through to LLM

[code-yeongyu]

Problem

claw permissions list, claw permissions allow bash, claw permissions deny bash etc. all fall through to the prompt/LLM path, calling the Anthropic API and returning auth errors. Only the bare single-word form claw permissions was intercepted by bare_slash_command_guidance.

Root Cause

parse_single_word_command_alias guards on rest.len() != 1 and returns None for multi-word forms, then parse_subcommand had no "permissions" arm so those forms hit _other => CliAction::Prompt.

Same root cause cluster as claw plugin list / claw marketplace fixed in #2993.

Fix

Add a "permissions" arm in parse_subcommand returning a structured guidance Err with the correct usage hint.

Verification

All 6 invocation forms tested — all exit 1 with kind:unknown guidance JSON, zero sessions created:

claw permissions              → guidance exit:1 ✅
claw permissions list         → guidance exit:1 ✅
claw permissions read-only    → guidance exit:1 ✅
claw permissions allow bash   → guidance exit:1 ✅
claw permissions deny bash    → guidance exit:1 ✅
claw permissions workspace-write → guidance exit:1 ✅