Add queue for multithread processing

[drew2a]

This is a continuation of the previous PR #7413 and the final fix for the deadlock problem.

It adds a thread-safe queue that is used for processing incoming MDS entities, which are created in a different thread.
Additionally, it removes unnecessary logic from Endpoints that perform writes to both databases (KnowledgeDB and MetadataDB).

As a drawback, this PR reduces the speed of tag extraction, as they are now not performed immediately upon entity addition to the MetadataDB. Given that tags aren't directly used in the current release and we plan to remove MDS in upcoming releases, this drawback is acceptable.

[kozlovsky]

It seems that with this change KnowledgeRulesProcessor performs better.

But process_batch still updates two databases - MetadataDB by updating torrent.tag_processor_version and KnowledgeDB by processing the torrent's title. In some next PR, it may be beneficial to modify process_batch so it starts adding titles to the queue instead of processing them directly; after that, updating both databases will be completely separated.

[drew2a]

But process_batch still updates two databases - MetadataDB by updating torrent.tag_processor_version and KnowledgeDB by processing the torrent's title.

Good catch, I overlooked that.
I've removed this line as it's not necessary. torrent.tag_processor_version should be set in all cases except process_batch (as it processes the MDB sequentially).