fix: resolve race conditions in Redis TTL and prevent cache test state leakage

[gr8man]

Description
This PR fixes persistent and random failures when running the cache test suite (tests/system/Cache) repeatedly or in a random order (--order-by=random).

Ref: #9968

Changes included:

1. Fix TTL Race Condition in Redis/Predis Handlers: Changed the PredisHandler and RedisHandler to use the expire() command directly with a relative TTL (in seconds) instead of calculating an absolute timestamp for expireAt() via PHP's Time::now(). Using a PHP-generated timestamp makes the cache vulnerable to clock drifts between the PHP server and the Redis server. This drift resulted in random test failures (e.g. testGet and testRemember failing to assert null after waiting for 3 seconds).
2. Prevent State Leakage in Tests: Several test classes (FileHandlerTest, RedisHandlerTest, PredisHandlerTest, ApcuHandlerTest, MemcachedHandlerTest) inherited testDeleteMatching which temporarily created 50-100 cache entries. The tearDown() methods previously only deleted three predefined default keys (key1, key2, key3), leaving dozens of dummy keys behind. This leaked state caused random Directory not empty system errors for FileHandler and false-positive assertion failures in tests like testDeleteMatchingNothing. tearDown() now properly calls $this->handler->clean() to guarantee a perfectly clean state for the next random test.
3. Proper ApcuHandlerTest Skip Logic: Added an explicit check in ApcuHandlerTest's setUp() to verify apcu_enabled(). This correctly skips the test when APCu CLI mode is disabled instead of falling back to the DummyHandler and producing cascaded test failures.

Checklist:

• Securely signed commits (if configured in your environment)
• Component(s) with PHPDoc blocks, only if necessary or adds value (without duplication)
• Unit testing, with >80% coverage
• User guide updated
• Conforms to style guide

[gr8man]

IMPORTANT

Note to reviewers and developers: Please ensure that tests are run regularly with the --order-by=random flag during development. Identifying test pollution and global state leakage bugs at a later stage is extremely difficult and time-consuming, as these failures only manifest under very specific test suite execution orders.