fix: remove stale session entry from sessions.json on idle-TTL eviction

[Stoltemberg]

What does this PR do?

When _sweep_idle_cached_agents() evicts an agent after the idle TTL (3600s), the session is ended in state.db but sessions.json was NOT updated. The stale session_key → session_id entry remained, routing all subsequent messages on that key into a closed session — silently dropping them until gateway restart (when _prune_stale_sessions_locked() fires).

Adds SessionStore.remove() method and calls it during idle eviction so sessions.json is cleaned up immediately.

Related Issue

Fixes #54878

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• gateway/session.py: Add SessionStore.remove(session_key) method that removes an entry from _entries and calls _save(), following the existing thread-safety pattern (lock + _ensure_loaded_locked)
• gateway/run.py: In _sweep_idle_cached_agents(), call self.session_store.remove(key) after popping the agent from cache, wrapped in try/except to avoid blocking eviction on a save failure

How to Test

1. Configure a gateway with a short idle TTL (or wait for the default 3600s)
2. Send a message, wait for idle eviction
3. Send another message on the same session key — should create a fresh session instead of being silently dropped

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this is not a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run ruff check gateway/session.py gateway/run.py and both pass
• I've tested on my platform: Windows

Documentation & Housekeeping

• I've updated relevant documentation — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — uses pathlib and threading, cross-platform safe
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

[alt-glitch]

This was generated by AI during triage.

Competing with #54916 for the same issue (#54878). Same goal — clear the stale sessions.json entry on idle-TTL eviction so messages aren't silently dropped — but a different mechanism: this PR adds SessionStore.remove() and calls it directly in _sweep_idle_cached_agents, while #54916 adds _cleanup_evicted_agent_session + remove_ended_session/_is_session_active validation. Related family: #55045 / #54293 fix the memory-provider on_session_end loss on the same eviction path. Maintainer should pick one of the two sessions.json fixes.

[tonydwb]

Code Review Summary

Verdict: LGTM

Clean fix — removes stale session entries from sessions.json after idle-TTL eviction to prevent messages from routing into ended sessions. The SessionStore.remove() method is well-implemented with proper locking and save.

Correctness: The remove() call in _sweep_idle_cached_agents is the correct injection point.
Testing: Dedicated test file with clear assertions for the remove operation.

---

Reviewed by Hermes Agent

[tonydwb]

Code Review Summary

Verdict: LGTM

Fixes stale session routing after idle-TTL eviction. Removes session entry from sessions.json when agent is evicted, preventing messages from routing into ended sessions.

• Clean fix with dedicated test
• Well-scoped: 3 files, 56 additions

Reviewed 3 files, 56 additions. Approved.

[Stoltemberg]

Closing — duplicate detected by alt-glitch bot. Consolidating into earlier PRs. 🙏