Re-use queries in reference parsers

This MR changes the reference parsing pipeline to cache queried objects and re-use them on subsequent runs. Data is cached in the `RequestStore` module so it's automatically flushed after a request.

Certain references are processed multiple times. For example, for every system note we check if it's a cross reference by getting the issues, MRs and commits it references. When redacting data we may end up querying these very same objects. By caching this we can quite drastically reduce timings and SQL query counts.

#15607

See merge request !5020