Add a Hash-based index with append-only B+ buckets

[leonardoeloy]

This one is a complete hunch I had a few months ago. I think we'll be able to circumvent the whole "stop everything, I'm compacting" situation if we store multiple append-only B+ trees.

Of course, this would lead to difficulties in file management and designing a proper hash function (preferably O(1)).

So, I'm thinking in developing this alongside with the traditional single-B+-tree approach. This way we can benchmark it and eliminate the hunch-oriented development.

Thoughts?