What does this PR do?

Fixes an aiohttp session leak in the Hindsight memory plugin that shows up as Unclosed client session / Unclosed connector errors in any long-running gateway hosting more than one concurrent chat.

Root cause. The module-global _loop / _loop_thread pair in plugins/memory/hindsight/__init__.py is shared across every HindsightMemoryProvider instance in the process. The plugin loader creates one provider per AIAgent (run_agent.py:1462), and the gateway creates one AIAgent per concurrent chat session (Telegram/Discord/Slack/WhatsApp/CLI). But HindsightMemoryProvider.shutdown() stopped the shared loop when any single session ended:

# Before (main)
if _loop is not None and _loop.is_running():
    _loop.call_soon_threadsafe(_loop.stop)
    _loop_thread.join(timeout=5.0)
    _loop = None
    _loop_thread = None

This stranded the aiohttp.ClientSession + TCPConnector owned by every sibling provider — their sessions were created on the loop (via hindsight_client_api.rest.RESTClientObject._ensure_session, which is one-session-per-client by design), and they cannot be closed once their loop is dead. They surface minutes later as the warnings reported in #11923.

Fix. Don't stop the shared loop on per-provider shutdown. Per-provider cleanup still closes that provider's own client via self._client.aclose() before self._client = None. The loop runs on a daemon thread and is reclaimed on process exit; keeping it alive between provider shutdowns lets sibling providers drain their own sessions cleanly.

Related Issue

Fixes #11923. Builds on (does not undo) the prior work in #4762 / #5094.

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

• plugins/memory/hindsight/__init__.py — HindsightMemoryProvider.shutdown() no longer stops the module-global event loop. Drops the now-unused global _loop, _loop_thread declaration. Replaces the loop-stop block with a comment explaining the sharing invariant.
• tests/plugins/memory/test_hindsight_provider.py — adds TestSharedEventLoopLifecycle with two regression tests:
  • test_shutdown_does_not_stop_shared_event_loop — primes the shared loop, creates two independent providers (two concurrent chat sessions), shuts down one, verifies the loop + thread are still the same live objects and that the second provider can still dispatch async work. Fails on main with the exact #11923 failure mode; passes with the fix.
  • test_client_aclose_called_on_cloud_mode_shutdown — confirms per-provider session cleanup still runs (mock_client.aclose.assert_called_once()).

How to Test

1. pytest tests/plugins/memory/test_hindsight_provider.py -q -o addopts= — 46 passed (44 existing + 2 new regression tests).
2. pytest tests/plugins/ -q -o addopts= — 221 passed.
3. pytest tests/ -q -o addopts= --ignore=tests/integration --ignore=tests/e2e --ignore=tests/acp --ignore=tests/tools/test_tts_kittentts.py — 14,109 passed (71 failed + 45 errored, all confined to tests/hermes_cli/test_web_server.py — an unrelated test file that also fails on pristine main; they do not touch plugins/memory/) (the excluded paths require optional-dep modules — acp, kittentts — not installed via pip install -e ".[dev]"; they also fail to collect on pristine main).
4. To see the regression test catch the bug on unfixed code:

   git revert HEAD -- plugins/memory/hindsight/__init__.py
   pytest tests/plugins/memory/test_hindsight_provider.py::TestSharedEventLoopLifecycle -q -o addopts=

   test_shutdown_does_not_stop_shared_event_loop fails with:

   AssertionError: shutdown() swapped out the shared event loop —
   sibling providers would have their aiohttp ClientSession orphaned (#11923)
   

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(hindsight): ...)
• I searched for existing PRs — no duplicate. (The only PR that matches the naive "aiohttp" search is fix(website): restore Docusaurus startup with webpack 5.106 #9043, a docusaurus/webpack website fix — unrelated.)
• My PR contains only changes related to this fix (no unrelated commits)
• I've run pytest tests/ -q and all tests pass (see "How to Test" above — full suite passes; the 8 optional-dep collection errors are pre-existing and unrelated)
• I've added tests for my changes (TestSharedEventLoopLifecycle)
• I've tested on my platform: macOS 15 / darwin-arm64 / Python 3.12.7

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — N/A (no user-visible config or behavior change beyond "the warning stops happening")
• I've updated cli-config.yaml.example if I added/changed config keys — N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
• I've considered cross-platform impact (Windows, macOS) — no platform-specific code touched; pure asyncio lifecycle change
• I've updated tool descriptions/schemas if I changed tool behavior — N/A