[Backend] Backfill changelog records from existing datasets

[cka-y]

Describe the problem

Once the pipeline is live, it captures only new changes. All consecutive dataset pairs already stored in the DB (and in GCS) represent a rich history of feed evolution that remains invisible.

Proposed solution

Add a backfill task to functions-python/tasks_executor/ that:

1. Queries all GtfsFeed records
2. For each feed, retrieves its GtfsDataset history ordered by downloaded_at
3. For each consecutive pair (previous, current) that doesn't already have a gtfs_dataset_changelog row (idempotency check): invokes the gtfs_diff module

The task should be rate-limited (e.g., N feeds per invocation) and restartable.

Alternatives considered

• Skip backfill entirely: acceptable if historical data is not needed at launch. The feature works forward-only. This issue can be deferred without blocking anything.