Bloom filter cache

[JonathanOppenheimer]

We can add a LRU cache for common.Hash to types.Bloom

OverrideHeaderBloom (rpc.go:295-301) reads raw receipts from DB and recomputes the bloom filter for every eth_getLogs filter scan. This is called per-block during log filter range queries, which means it could happen a ton of time.

There is no libevm equivalent because libevm stores blooms in headers directly.

The cache could be 2,048 entries (~512 KB — each Bloom is 256 bytes), and we could populate it when the blooms are computed and when blocks settle.