[luthermonson]

What changed

Backfills a real flush() now that ePHPm v0.1.2 ships the ephpm_kv_flush_all(): bool SAPI function.

• KV-ops trio — KvOpsInterface, SapiKvOps, InMemoryKvOps all gain flush(). SapiKvOps::flush() calls ephpm_kv_flush_all() behind a function_exists guard so older runtimes return false instead of fataling; InMemoryKvOps::flush() clears its $values/$deadlines arrays.
• Framework wiring — EphpmStore::flush() previously hard-returned false (no key enumeration was possible). It now delegates to $this->ops->flush(), so Cache::flush() and php artisan cache:clear actually clear the store. It still returns false on pre-v0.1.2 runtimes via the guard.
• Guarded for pre-v0.1.2 — on runtimes without the SAPI function, flush is a no-op returning false (recommended fallback: bump the cache prefix and let TTL age entries out).

Test

• New test_flush_clears_all_entries (uses InMemoryKvOps): puts two keys, flushes, asserts both gone and flush() returned true.
• Full suite: OK (28 tests, 75 assertions) under PHP 8.3. composer validate --strict passes.

Incidental fixes (required to make the suite run)

• The service provider registered the cache driver in register(), calling the Cache facade before the framework's cache manager was bound — this threw Target class [cache] does not exist and broke the entire suite (CI on main is currently red for this reason). Moved the Cache::extend() call to boot(), the standard spot for driver extensions.
• Fixed two pre-existing test bugs that surfaced once the suite executed: an assertion comparing serialized bytes to the logical value, and a call to the non-existent CacheManager::getDriver().

Do not merge yet.