feat(locations): add a cancel hook to the endpoints-to-locations backfill

[Maffooch]

[sc-14596]

What

Adds a cooperative cancel_callback stealth option to migrate_endpoints_to_locations, so an in-process caller can stop a long backfill cleanly at the next chunk boundary. This is the OSS half of adding Cancel + re-run to the Locations data-migration suite; the user-facing button lives in Pro and consumes this hook.

Why

The endpoints backfill can run for a long time, and reportedly gets stuck with no way to stop it. There was no cancellation path in the command at all. Because each chunk is separately committed and the command is idempotent (it converges on re-run), stopping between chunks is safe and resumable.

How

• cancel_callback() -> bool added to stealth_options and read in handle(), alongside the existing progress_callback / summary_callback.
• _should_cancel() is polled at the chunk boundary, right after _emit_progress. A raising probe is treated as "keep going" (a transient error must not throw away hours of work), matching the existing swallow-and-continue convention for the other hooks.
• On cancel the run breaks out of the loop but still runs the tag-inheritance pass for what it migrated (it's idempotent), and reports {"cancelled": True} in the summary so the caller can land the run as cancelled rather than completed.
• CLI runs pass no callback and behave exactly as before.
• Docs (PRO__migrating_from_endpoints, PRO__feature_flags) and the Pro changelog (v3.2.300) describe the Cancel behaviour.

Ordering

This must merge before the paired Pro PR: the Pro migration suite passes cancel_callback into call_command, which raises on an unknown stealth option until this ships in the image the Pro build consumes.

Testing

./run-unittest.sh --test-case unittests.test_migrate_endpoints_to_locations — 15/15 pass (4 new: cancel-between-chunks, re-run converges, raising probe doesn't abort, CLI path unchanged). manage.py check and makemigrations --check clean; ruff clean.

🤖 Generated with Claude Code