fix(tui): keep background notifications in owning session (#42674)

[raywcm]

What does this PR do?

Fixes a Desktop/TUI cross-session notification path for background process completions.

The poller path already skips events whose session_key belongs to another live session, but the post-turn safety drain still called process_registry.drain_notifications() directly. That method drains the global process completion queue, so a turn finishing in session B could consume and dispatch a notify_on_complete event that was started by session A.

This PR adds a small TUI-owned drain helper that filters drained events with the same ownership check used by the poller, requeues foreign live-session events, and only dispatches notifications owned by the current session (plus global/orphan notifications).

Related Issue

Fixes #42674

Related to #35652, but this targets the post-turn drain path rather than the between-turn poller path.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• tui_gateway/server.py
  • Added _drain_owned_notifications() to drain only notifications owned by the current TUI session.
  • Requeues foreign live-session events so the owning session's poller can handle them.
  • Uses the helper in the post-turn completion-notification drain.
• tests/test_tui_gateway_server.py
  • Added regression coverage for owned, foreign, and global notification events.

How to Test

1. Open two TUI/Desktop sessions.
2. In session A, start a bounded background process with notify_on_complete=True.
3. While it is running, send a normal message in session B.
4. When the process completes, the notification should remain routed to session A instead of being consumed by session B's post-turn drain.

Local verification performed:

python3 -m py_compile /tmp/hermes-fix/tui_gateway/server.py
python3 -m py_compile /tmp/hermes-fix/tests/test_tui_gateway_server.py

I could not run the full pytest suite in this PRoot environment because cloning/extracting the full repository repeatedly timed out; the patch was prepared via GitHub Contents API against main and syntax-checked locally.

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 isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: Linux / PRoot (syntax checks)

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — 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 — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

For New Skills

N/A

Screenshots / Logs

Syntax checks:

python3 -m py_compile /tmp/hermes-fix/tui_gateway/server.py
python3 -m py_compile /tmp/hermes-fix/tests/test_tui_gateway_server.py

[teknium1]

Thanks for isolating the post-turn drain path. The premise is still valid on current main: tui_gateway/server.py:9408 invokes the post-turn drain, while tools/process_registry.py:1200 applies ownership filtering only to async_delegation, so ordinary completion events still bypass the callback.

Problems

• The patch predates the current ownership API: current tui_gateway/server.py:8545 defines _notification_event_belongs_elsewhere(sid, session, evt), while the proposed helper calls it with two arguments.
• Filtering after drain_notifications() is too late for events suppressed inside that method: tools/process_registry.py:1195 runs completion suppression before the proposed helper would receive an event.
• The added test directly calls the helper rather than exercising _run_prompt_submit() with a foreign ordinary completion.

Suggested changes

• Put addressed-event ownership routing in ProcessRegistry.drain_notifications() before completion suppression, using the current compression-aware _session_owns_notification_event() callback from the TUI.
• Add real post-turn A/B tests for successful and failed completions, plus compression-lineage ownership.

Automated hermes-sweeper review.

[teknium1]

Closing as superseded by #65419 (merged) — the post-turn drain now passes session_key + the compression-chain-aware owns_event callback + skip_poll_observed=False, which covers everything _drain_owned_notifications() did plus lineage resolution and a requeue-when-running guard. Your diagnosis of the bare drain_notifications() call was correct a month before the rest of us got there — thanks @raywcm, and sorry this sat long enough to be fixed out from under it.