fix: Fix cache problems with lints level

[ChayimFriedman2]

Fixes #19823.

I knew this cache will cause us problems.

I question the need for a cache (yep, I authored this code, I know). When there is no cache there are no cache problems. And we all know what the hardest problems in computer science are. This code was just needlessly complex.

Ideally I believe we would have lint levels defined as db queries and query them before firing the lints. Unfortunately in rustc lint levels are tied with spans, which can be arbitrary pieces of code, and we can't implement something like that for our hir. So we're stuck with AST-based lint resolution (which isn't that bad, actually).

[Veykril]

Given diagnostics should be rare (aside from syntax errors), in that the are only a handful usually, the cache is likely not going to be that impactful either way.

[ChayimFriedman2]

My original thought was that some files, e.g. binding files, have some lints (case lints) triggered on large amounts. That is correct, but now I don't think it justifies the code complexity.