Fix for cyclemanager not resetting unregisterRequested flag

[aliszka]

What's being changed:

Addresses: #4418

Callback registered in cyclecallbackgroup can be unregistered or deactivated. Both of this actions used to overwrite shouldAbort function passed to callback to always return positive value, indicating that long-running callback should abort its execution whenever possible. Overwritten shouldAbort function was never restored after callback was activated. As a result callback once deactivated was never successfully activated again.

To ensure all data already sent to database are included into backup, they have to persisted in files. Therefore memtables are forced to flush. Forced flush requires deactivating and activating back again background flush process responsible for flushing dirty memtables in intervals.
Due to bug described above, background flush process was not correctly resumed, resulting in data being stored in memory and wals until db shutdown (forced flush) or db restart (wal ingestion).

This fix makes sure shouldAbort function is restored to its original form when callback is activated after being deactivated.

Review checklist

• Documentation has been updated, if necessary. Link to changed documentation:
• Chaos pipeline run or not necessary. Link to pipeline: [TEST] cyclemanager fix weaviate-chaos-engineering#186
• All new code is covered by tests where it is reasonable.
• Performance tests have been run or not necessary.

[sonarcloud]

Quality Gate failed

Failed conditions
66.8% Duplication on New Code (required ≤ 3%)

See analysis details on SonarCloud