Improve linkage cache linkage_cache_performance

[MikeMcQuaid]

• Use reference counting so nested CacheStoreDatabase.use share the same underlying database (rather than rereading it every time).
• Only write out the cache database file when it has changed.
• Cleanup cache database entries on formula or full brew cleanup.

Fixes #8690

• Have you followed the guidelines in our Contributing document?
• Have you checked to ensure there aren't other open Pull Requests for the same change?
• Have you added an explanation of what your changes do and why you'd like us to include them?
• Have you written new tests for your changes? Here's an example.
• Have you successfully run brew style with your changes locally?
• Have you successfully run brew tests with your changes locally?

---

[mistydemeo]

With this branch, performance of brew install improved to about 45 seconds total; we shaved off 1/3 of the total time. brew cleanup shrunk the linkage cache to under 500KB, so that's looking like a good longterm solution here. Notably, that's significantly larger than the size I'd get from deleting the linkage cache entirely and letting it be recreated, but at this size we're spending at most a second in the linkage checker - much more reasonable.

I do notice that the performance at this stage is still moderately worse than having no cache at all - I wonder if we've got more gains to be made. It could also be that the way we're handling this means that we have performance gains from caching with small numbers of formulae installed, but at certain cache sizes performance ends up actually being better without a cache.

[claui]

SCNR @MikeMcQuaid

[claui]

Thank you @MikeMcQuaid. That performance gain is immensely helpful!

[MikeMcQuaid]

Thanks for testing @mistydemeo!

I do notice that the performance at this stage is still moderately worse than having no cache at all - I wonder if we've got more gains to be made. It could also be that the way we're handling this means that we have performance gains from caching with small numbers of formulae installed, but at certain cache sizes performance ends up actually being better without a cache.

Yeh, there's definitely improvement to be had here although it's likely very dependent on your disk IO (doing this on an spinning platter HDD will take a lot longer to be slower, from previous measurements). The reference counting I've added here should mean you can wrap multiple calls to e.g. undeclared_runtime_dependencies with a top-level CacheStoreDatabase.use(:linkage) which should avoid reading the file from disk multiple times when not necessary.