Reduce DB size with bloom filter

[gwillem]

Have ran some benchmarks, a Binary Fuse filter (0.3% FPs) is the fastest (6ns bf lookup vs 14ns per map lookup) and smallest (~25% of orig []uint64 size) with our dimension data.

2022 article: http://arxiv.org/abs/2201.01174
Go Library: https://github.com/FastFilter/xorfilter

Benefit: we don't have to rebuild hashes, as we can populate it with the existing hash db.
However, corediff now needs to deal with two types of databases, of which one is immutable. Requires some refactoring of the checkPath function.