[Bug] /api/v1/items/reindex endpoint missing session.commit() — new episodes never persisted

[bioengineered]

Summary

The /api/v1/items/reindex endpoint successfully fetches new episodes from TVDB and merges them into the session, but never calls session.commit(). The session rolls back on close, so new seasons/episodes are silently lost. The endpoint returns "Successfully re-indexed" even though nothing was persisted.

Steps to Reproduce

1. Have a show with new episodes available on TVDB that aren't in the Riven database yet
2. Call POST /api/v1/items/reindex with the show's item_id
3. API returns {"message": "Successfully re-indexed <show>"}
4. Check the database — no new episodes were added

Root Cause

In src/routers/secure/items.py, the reindex endpoint calls apply_item_mutation() but never follows it with session.commit(). Every other caller of apply_item_mutation in the same file does call session.commit() immediately after. The function's own docstring states: "Caller is responsible for session.commit()."

The scheduler's _run_reindex_for_item performs the same operation and does commit — so the scheduler path works correctly. Only the API endpoint is affected.

Relevant Code

src/routers/secure/items.py — the reindex handler (around line 1356):

            apply_item_mutation(
                program=di[Program],
                session=session,
                item=item,
                mutation_fn=mutation,
                bubble_parents=True,
            )

            # <-- session.commit() missing here

            logger.info(f"Successfully re-indexed {item.log_string}")

Compare with every other apply_item_mutation caller in the same file, which all have session.commit() after.

Fix

Add session.commit() between apply_item_mutation(...) and the logger.info(...) line, matching the pattern used by all other callers.

Environment

• Riven v1.0.0 (spoked/riven:dev, image built 2026-04-11)
• PostgreSQL 17 Alpine

[bioengineered]

Additional testing on spoked/riven:dev (06f5a78): the assumption here that the scheduler path persists correctly because it commits doesn't hold. _run_reindex_for_item does commit, but new episodes still aren't saved — they're never enlisted in the session to begin with.

_create_episode_from_data sets episode.parent = season before Season.add_episode runs. The back-populate appends the transient episode to season.episodes, but a transient child attached through the many-to-one backref isn't picked up by the save-update cascade when the season was loaded via show.seasons. add_episode then sees the number already present and returns early, so session.add() is never called.

Verified after indexer.run(show): inspect(new_ep).transient is True, new_ep not in session, and flush/commit inserts nothing. Adding the missing session.commit() here is still correct, but it won't persist new episodes on its own — the enlistment issue in Season.add_episode has to be fixed as well. Root cause and a fix in #1398.