Use the redis set cache for branch and tag names

What does this MR do?

This completes the re-introduction of gitlab-foss!30476 (merged), which was reverted in gitlab-foss!32408 (merged) and partially reintroduced in gitlab-foss!32412 (merged)

The revert was because the code didn't handle a HA environment with mixed code well - incorrect cache invalidation meant that sometimes branches or tags would seem to be missing.

To resolve this, we introduced an abstract "redis set cache" into %12.3 and started sending cache-invalidation messages to it. This means that in %12.4, we can start using that cache for branch names and tag names.

On git push, or any other event that changes the list of branches and tags, 12.3 will invalidate both the old Rails.cache-based cache, and the new redis set cache, and so will 12.4. 12.2 only invalidated the former.

On read, 12.3 will only read from the former; 12.4 will only read from the latter.

In a mixed-code environment, we'll hold the values in Redis twice, but each will always be correct. Once the mixed-code environment is fully upgraded to 12.4, the duplicate copies will go away.

Using a redis set allows us to check whether a branch exists using the SISMEMBER Redis command, instead of reading the full list of branches from Redis then doing an array membership check. This is a big improvement in some circumstances.

Screenshots

Does this MR meet the acceptance criteria?

Conformity

• Changelog entry
• [-] Documentation created/updated or follow-up review issue created
• Code review guidelines
• Merge request performance guidelines
• Style guides
• [-] Database guides
• Separation of EE specific content

Closes #30439 (closed)