fix(web): optimize pending send pruning for YEF-899

[Yevanchen]

Summary

• Optimized pending-send pruning by indexing user message ids by trimmed content before reconciling optimistic sends.
• Added coverage for the case where matching text includes both a baseline message and a new server echo.

Why

• Long agent sessions can accumulate many messages; the old pruning path scanned every message for every pending send and did a linear baseline-id check inside that scan.
• The new path changes reconciliation from roughly O(pending * messages * baseline ids) to O(messages + pending * (baseline ids + matching messages)) while preserving trimmed-content and non-baseline-id semantics.
• Closes YEF-899

Verification

• Commands:
  • just test-file apps/web/tests/agent-session-pending-sends.test.ts
  • just tc-package @mosoo/web
  • just fmt-check-path apps/web/src/routes/agent/components/agent-session-pending-sends.ts
  • just fmt-check-path apps/web/tests/agent-session-pending-sends.test.ts
  • just test-package @mosoo/web
  • just check
• Manual steps: N/A
• Not run: N/A

Impact

• User/API/contract changes: No API or contract changes; optimistic send cleanup is faster for long sessions.
• Generated files / GraphQL / DB / lockfile: None.
• Env or config changes: None.
• Risk and rollback: Low; rollback is reverting the model and test commit.

Review

• Closest review areas: pending-send reconciliation in agent-session-pending-sends.ts.
• Known trade-offs: Each prune sweep now builds a linear message index, replacing repeated full-message scans.