fix: linearize scheduled execution cancel with publish - #14225

What does this PR do?

Scheduled execution cancellation raced with the functions worker's claim, queue publication, and cleanup.

1. The worker claimed a schedule (active=false). A concurrent cancel saw that claim and was rejected. If publication then failed, the worker restored active=true, so the execution could still run.
2. If publication succeeded but schedule cleanup failed, cancelling the inactive schedule could still report success even though the function job was already queued.

Publication and cancellation now share one coordinator, scheduleForExecutions, registered as an object dependency. It holds lock:platform:schedules:{scheduleId} across claim, publish, cleanup, failed-publish restore, and cancel.

• 204 only when the schedule is still active under that lock, so publication was prevented.
• A cancel that loses to a successful publish is rejected as execution_in_progress (400), including when cleanup left the row inactive.
• A failed publish restores active=true before it releases the lock, so a cancel that was waiting deletes the schedule instead of leaving it runnable.
• If the lock cannot be acquired within 10 seconds, cancel throws general_resource_locked (409). The schedule is left unchanged and the client can retry. A timeout is not treated as losing to publication, so a later failed publish cannot revive a schedule after a rejected cancel.

Fixes #13249

Test Plan

• php vendor/bin/phpunit tests/unit/Schedule/ExecutionTest.php tests/unit/Platform/Workers/FunctionsTest.php --no-configuration --bootstrap app/init.php (10 tests, 79 assertions)
• php vendor/bin/phpstan analyse on the changed files
• php vendor/bin/pint --test on the changed files

Library tests in tests/unit/Schedule/ExecutionTest.php wait on the lock with fibers and assert both the cancellation result and persistence:

• Cancel before publish, so the worker does not enqueue and cancel() returns true.
• Cancel that waits through a failed publish: cancel() returns true and the restored row is deleted.
• Cancel that waits through a successful publish whose cleanup fails: cancel() returns false and the inactive row stays.
• Cancel of an already claimed schedule does not delete it.
• Lock acquisition timeout throws general_resource_locked and leaves the active schedule in place.

The existing e2e testDeleteScheduledExecution remains the HTTP contract for a cancel that wins outright.

Related PRs and Issues

• Fixes Scheduled execution cancellation can race with queue publication #13249

Checklist

• Have you read the Contributing Guidelines on issues?
• If the PR includes a change to an API's metadata (desc, label, params, etc.), does it also include updated API specs and example docs?