fix(sessions): release unused SQLAlchemy table-init locks

[rioyu123]

Summary

This pull request fixes unused SQLAlchemy table-initialization locks remaining in the process-wide registry.

SQLAlchemySession(create_tables=True) currently keeps one initialization lock for every distinct database/table configuration, even after all sessions using that configuration have been released.

The registry now holds weak values, while each live session keeps its existing strong lock reference. Sessions sharing a key still use the same lock; the registry guard, initialization ordering and cancellation handling are unchanged.

In a probe creating and discarding sessions for 100,000 distinct configurations, the old registry retained 100,000 entries after collection. The updated registry retained none. The measured allocation delta came from tracemalloc; it is not RSS or the byte size of the lock objects.

Test plan

• Added regression coverage for collection of unused locks and sharing with surviving owners during concurrent first writes.
• Confirmed the collection regression fails against the unchanged base and existing cross-loop and cancelled-waiter tests still pass.
• Ran the complete format, lint, Mypy, Pyright and test verification on Python3.12, plus focused tests on Python3.10.

Checks

• Added regression tests.
• Ran .agents/skills/code-change-verification/scripts/run.sh.
• Confirmed all verification steps pass.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-08-29T03:49:44.768056Z	f4cc0db	PR opened
🔒 Security Review	✅ Completed 2026-08-29T03:50:54.307779Z	f4cc0db	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.