[CHIA-1608] Use a cache eviction logic that matches our BlsCache usage pattern

[AmineKhaldi]

No description provided.

[coveralls-official]

Pull Request Test Coverage Report for Build 11854445955

Details

• 88 of 100 (88.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.006%) to 83.862%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
crates/chia-bls/src/bls_cache.rs	88	100	88.0%

Totals
Change from base Build 11851422401:	0.006%
Covered Lines:	12913
Relevant Lines:	15398

---

💛 - Coveralls

[arvidn]

did you run the benchmark to compare against main? I think that would be an interesting comparison.

Could you also make sure that it's OK to evict on a cache hit? i.e. that we never rely on two or more cache hits?

If there's a reorg we might get more cache hits, but we should probably not optimize for that case. If there are competing unfinished blocks is another example where we'll have multiple cache hits.

It might make sense to make the eviction explicit. When a block is fully validated and added to the chain it should be safe to evict its keys from the cache.