MRG: Add threaded multigather to pyo3_branchwater code; add fastmultigather CLI plugin.

[ctb]

Add fastmultigather to plugin.

This PR refactors the code into more reusable chunks and adds fastmultigather, which will take in multiple queries and run gather upon them. The basic memory usage will increase from fastgather because the multigather code loads multiple metagenomes into memory, but in exchange we only load the database once (at the beginning) and then process multiple metagenomes with it. 🎉

Specifically, the multigather code does the following:

• load all database sketches into memory;
• then, in a threaded loop,
  • load each query metagenome
  • do a threaded prefetch to find matching sketches
  • follows that with a gather (threaded prefetch + pick best + subtract, repeat)