fix: clean up orphaned messages in delete_branch and propagate add_items metadata errors

[kimnamu]

Summary

Two correctness fixes for AdvancedSQLiteSession, each with a regression test.

#3346 — delete_branch() removes message_structure rows but never cleans up the
actual messages, leaking storage indefinitely. The fix reuses the existing
_cleanup_orphaned_messages_sync() (which add_items() already calls on failure)
so there's no new deletion logic — just a missing call site.

#3348 — add_items() catches _insert_structure_metadata() failures, runs cleanup,
but then returns normally. Callers cannot distinguish success from partial failure.
The fix adds a raise RuntimeError(...) from e after the existing cleanup block (3 lines).

Approach

Both fixes reuse existing infrastructure rather than introducing new code:

Fix	Approach	Why
#3346	Call self._cleanup_orphaned_messages_sync(conn) after structure deletion	This method already exists at L491 and uses the same LEFT JOIN ... IS NULL pattern — no new query logic needed
#3348	Add raise RuntimeError(...) from e after the existing except cleanup block	The cleanup + logging already exists; only the re-raise was missing

Before / After

Scenario	Before (main)	After (this PR)
delete_branch("feat") with branch-only messages	message_structure deleted, orphan rows stay in messages_table forever	Orphans removed via existing _cleanup_orphaned_messages_sync()
add_items([...]) when metadata fails	Logs error, returns None — caller assumes success	Raises RuntimeError with cause chain; caller can handle or retry
Run Test
Existing test suite	36 pass	36 pass (zero regressions) + 2 new

Test plan

• All 36 pre-existing tests pass unchanged (zero regressions)
• test_delete_branch_cleans_up_orphaned_messages: creates branch-only messages, deletes branch, asserts messages_table is clean
• test_add_items_raises_on_structure_metadata_failure: monkeypatches metadata insertion to fail, asserts RuntimeError and verifies orphan cleanup ran

$ uv run pytest tests/extensions/memory/test_advanced_sqlite_session.py -v
...
38 passed in 0.22s

Issue number

Closes #3346
Closes #3348

Checks

• I've added new tests (if relevant)
• I've added/updated the relevant documentation
• I've run make lint and make format
• I've made sure tests pass

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 659f368812

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[kimnamu]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. More of your lovely PRs please.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[kimnamu]

Closing in favor of two focused, single-issue PRs to make review easier:

• fix: #3346 clean up orphaned messages in delete_branch #3526 — AdvancedSQLiteSession.delete_branch() leaves branch-only messages in the base table #3346 orphaned message cleanup in delete_branch
• fix: #3348 propagate add_items structure metadata failure #3527 — AdvancedSQLiteSession.add_items can report success after structure metadata failure #3348 propagate add_items structure metadata failures

Both reuse the session's existing helpers and each ships with its own regression test. The AdvancedSQLiteSession internals (especially the shared _cleanup_orphaned_messages_sync helper) made these fixes very clean to slot in — thanks for the well-structured code. Happy to adjust anything on the new PRs.