[jeremystretch]

Closes: #23463

Queued object events now carry the IDs of the change log records behind them, so webhooks and EVENTS_PIPELINE consumers can tie an event to its changelog entries.

• The change logging signal handlers pass the ID of the ObjectChange they record (or update, for M2M changes) to enqueue_event(), which takes a new optional object_change_id argument.
• Each queued event gets an object_change_ids list. Changes to the same object within a request are coalesced into one event, so the list can hold several IDs, in the order they were recorded and without duplicates. A save that records no change adds nothing.
• The webhook action passes the list through to send_webhook(), which adds object_change_ids to the webhook context when it's not empty, the same way snapshots is handled. Job events never have one.
• Updated the webhook docs (available context, default body example, and the context table).

This follows the implementation proposed in the issue, plus the pass-through to webhooks (the issue's diff stops at the event queue). The list is created when an event is first queued, instead of using setdefault() when events are coalesced.

The change is additive: enqueue_event() and send_webhook() only gain optional arguments, and webhook jobs already queued before an upgrade still run. The reverse isn't fully covered: a worker still running older code will reject a webhook job that carries object_change_ids. To keep that window small, the argument is only passed when there are IDs to send.

Tests: extended the create/update/delete event rule tests to check the IDs against the recorded ObjectChange. Added tests for a coalesced event (create, update, M2M change and a no-op save in one request), for the webhook body, and for omitting the key when there are no IDs. I also checked it end to end against a scratch instance: created, updated, no-op-updated and deleted a site through the REST API, and in each webhook received, object_change_ids matched the changelog entry for the same request (the no-op update had none). The full test suite, ruff check, makemigrations --check and the docs build all pass.

[pheus]

Thanks! This looks good to me.