feat: odpt integration

[ianktc]

Summary:

odpt_script as provided directly from ODPT

This PR adds a new odpt_import task to the tasks_executor Cloud Function that imports Japanese transit feeds from the Open Data Platform for Transportation (ODPT). It queries ODPT's members-portal metadata API for openly-licensed (CC BY 4.0 and CC0) GTFS datasets, creates/updates the corresponding Gtfsfeed and Gtfsrealtimefeed (trip updates, vehicle positions, service alerts) records, tags each with an odpt external ID and a Japan-level Location, and links related GTFS-RT feeds back to their parent schedule feed. Closes MobilityData/mobility-database-catalogs#1391. A new monthly Cloud Scheduler job (odpt_import_schedule, 5th of the month) is added, alongside the existing JBDA/TDG import schedules.

Expected behavior:

• Running the task with dry_run: true (default) fetches and diffs against the DB but performs no writes — all staged session changes are explicitly rolled back at the end.
• Running with dry_run: false creates new GTFS/GTFS-RT feeds for previously-unseen ODPT org/dataset pairs, updates existing ones only when a tracked field (name, provider, producer URL, license URL, or an RT endpoint URL) actually changed, and leaves unchanged feeds untouched.
• New/changed feeds trigger a dataset-download Pub/Sub message; changed feeds also enqueue a website cache-revalidation task.
• If a schedule feed's producer URL changes in place, a feed.url_updated notification event is emitted.
• Changes are committed in batches (COMMIT_BATCH_SIZE, default 5) rather than one giant transaction, with per-batch IntegrityError handling that rolls back just that batch.

Testing tips:

• Run scripts/api-tests.sh --folder functions-python/tasks_executor — includes the new test_odpt_import.py covering: full create path (schedule + 3 RT feeds + location + external ID), the dry-run-must-not-write regression test, and graceful handling of an upstream HTTP failure.
• To exercise locally end-to-end, invoke the tasks_executor function with {"task": "odpt_import", "payload": {"dry_run": true}} first and inspect the returned summary (created_gtfs, updated_gtfs, created_gtfs_rt, updated_gtfs_rt, linked_refs, total_processed_items) before flipping to dry_run: false against a local DB.

Please make sure these boxes are checked before submitting your pull request - thanks!

• Run the unit tests with ./scripts/api-tests.sh to make sure you didn't break anything
• Add or update any needed documentation to the repo
• Format the title like "feat: [new feature short description]". Title must follow the Conventional Commit Specification(https://www.conventionalcommits.org/en/v1.0.0/).
• Linked all relevant issues
• Include screenshot(s) showing how this pull request works and fixes the issue(s)

[ianktc]

I found the same issue with @with_db_session annotation committing on dry_run=True without an explicit rollback. This is addressed by adding an explicit rollback. if not dry_run then commit else rollback. I believe the same issue exists in tdg and jbda imports.

Explanation:

1. _import_odpt(dry_run=dry_run) is called with no db_session kwarg (line 88 of import_odpt_handler).
2. with_db_session's wrapper sees kwargs.get("db_session") is None, so it creates a Database instance and opens db.start_db_session() itself, injecting the new session into kwargs["db_session"].
3. start_db_session (lines 242–250) is a plain try/except: yield session (runs _import_odpt's whole body) → session.commit() if nothing raised → session.rollback() only if an exception propagated out.
4. _import_odpt returns a normal dict in the dry-run case (it never raises an exception) so branch is always "commit," never "rollback."