[ogorzalka]

Raises the coverage of the package from 84.7 % to 97.5 % (measured locally with PHP 8.4 and pcov; equal to the recorded ratchet), with tests only (src/ is untouched).

40 tests (107 to 147) in five new files and a fixture: the cron-event and migrate commands, the registrar, the edges of the engine (the cron options interceptor, CronIndex), and the schedule surface (fluent shortcuts, ScheduleFake, ScheduleDiscovery). tests/Support/WpCliFake.php simulates WP-CLI (WP_CLI, Cron_Event_Command, Formatter, WP_Error), each guarded by class_exists. Time is fixed: ScheduleEngine has an injectable clock (useClock) and the tests use fixed timestamps; only ActionSchedulerDriver reads time().

One existing test is touched: PackageBoundaryTest gets six lines. Once the WP-CLI fakes are loaded, the architecture test sees WP_CLI, WP_CLI\Formatter, Cron_Event_Command, is_wp_error and wp_list_pluck used in src/, which it ignored before only because they did not resolve; they are added to its ignoring list with the other WordPress names. Nothing else in that test is relaxed.

Not covered, and why: the ActionScheduler-absent paths of ActionSchedulerDriver (lines 50, 58; the fake class stays defined for the process), Schedule.php 162-164 (driver choice when add_filter does not exist: Brain Monkey defines it for the process), the AssertionError fallback of ScheduleFake without PHPUnit, the Cron_Event_Command-absent branch of Registrar (Pest 3 has no runInSeparateProcess()), and a few isolated lines. Seen in the code and left as it is: ScheduleDiscovery.php:135 builds the class name for a class without namespace as \Name, so a log message starts with a backslash (cosmetic).

The ratchet in .harness.json goes from 84.7 to 97. The 147 tests pass in random orders and in a Linux container (php:8.4-cli); Pint, PHPStan, Rector (--dry-run) and tests/Arch.php are clean.