implementing low-memory prefetch for one-vs-many comparisons (resolves #27)

[mahmudhera]

• remove stale files
• use proper threading in index load (previously it was using as many threads as there are chunk files in the stored index)
• modularize extraction of index
• modularize index info read (reading the summary file)
• implement single threaded prefetch
• add python test to ensure sourmash prefect produces the same results
• add sourmash prefetch outputs in repo
• add tests against the outputs in ci.yml