Bugfix: Implement transaction locks in cache invalidate

[MGApcDev]

Sometimes stale cache never purged in distributed system (race condition)

In bigger systems with a lot of traffic there is a chance that a cache is stuck in the stale state.

---

Bug scenario:

1. User performs query that triggers the invalidation of a model cache on "tag2"
2. Process 1; fetches hashes for a tag "tag2"
3. Another user performs query that triggers caching of data.
4. Process 2; adds cache for "key3",
5. Process 2; adds additional key "key3" to tag "tag2"
6. Process 1; deletes all hashes it found in tag "tag2"
7. Process 1; deletes cache tag "tag2"

Expected outcome:

• There should be a cache for "key3"
• There should be a cache for "tag2" containing "key3"

Actual outcome:

• There is a cache for "key3"
• There is NOT a cache for "tag2"

Issue:

• This is a problem, because the cache for "key3" will still be used by fetch queries, but going forward calls to invalidate "tag2" does not clear "key3".
• "key3" cache will say until the cache hits it's expire time.

---

I've tried to add a test that simulates the issue case, but it happens between 2 private method calls, so it's a little hard to write.

[spiritix]

Thanks for your contribution, highly appreciated!