Fix race condition in locks issued by the file cache driver

[hbgl]

This pull request fixes issue #43627.

TLDR:
Atomic locks issued by FileStore or other stores that use the HasCacheLock trait have a race condition when the lock timeout is zero (the default). This can cause multiple processes to concurrently enter a critical section protected by the lock.

The old lock implementation would do a non-atomic get and put when the lock timeout is less than or equal to zero and an atomic add otherwise. The new implementation always does an atomic add. This change only breaks code that incorrectly implemented their cache store, i.e. their add method cannot handle lock timeouts less or equal to zero.

[taylorotwell]

Not sure I'm comfortable changing this on a patch release.

[driesvints]

@hbgl can you try sending to master?

[hbgl]

@driesvints Here is the new PR for master #43686