[tests] Run the distributed locks on fakeredis

[nledez]

Problem

• The CI provides no Redis, and redis_lock.get_redis_client opens a fresh client and pings it on every lock
• Each playlist, annotation or cache single-flight lock taken in a test therefore waited for redis-py's connection retries (about ten seconds) before falling back to the unlocked degraded mode: tests/blueprints/files/test_extract_annotated_frame.py
  alone took 112 s instead of 2 s
• The locks themselves were never exercised by the suite, since every test ran in the degraded mode where a lock is always granted

Solution

• Point redis_lock.get_redis_client at a fakeredis in the test base, flushed after each test like the revoked token, config and job queue stores already are
• Install fakeredis with its lua extra (fakeredis[lua]==2.35.1, pulls lupa): redis-py's Lock releases through a Lua script, which fakeredis only runs with it
• Add tests/stores/test_redis_lock.py covering the lock exclusivity until release and the independence of two different keys
• Without a Redis server, the full suite drops from 22 minutes to under 10 locally, and the locks are now really tested