[filter] Index input sequences and stream output sequences

[huddlej]

Description of proposed changes

Reduces memory needed by augur filter to a smaller constant value
instead of loading all input sequences into memory once and then again
when writing sequences out to disk. The primary improvements here are:

1. Use a BioPython index data structure [1] that tracks where each
   sequence is on disk but does not load sequences into memory. This
   structure acts the same as the dictionary structure we originally used
   except sequences are loaded lazily when they are requested.

2. Use an iterator to write sequences back to disk. BioPython's SeqIO
   write method accepts an iterator that allows us to stream sequences back
   to disk without first loading them into memory as a list of sequence
   objects.

[1] http://biopython.org/DIST/docs/tutorial/Tutorial.html#sec66

Testing

This code has been tested with a recent ncov dataset to confirm the same number of sequences are output for a given input and filters. It also passes all functional and unit tests.

[codecov]

Codecov Report

Merging #627 into master will increase coverage by 0.02%.
The diff coverage is 83.33%.

@@            Coverage Diff             @@
##           master     #627      +/-   ##
==========================================
+ Coverage   28.53%   28.56%   +0.02%     
==========================================
  Files          39       39              
  Lines        5365     5367       +2     
  Branches     1319     1320       +1     
==========================================
+ Hits         1531     1533       +2     
  Misses       3778     3778              
  Partials       56       56              

Impacted Files	Coverage Δ
augur/filter.py	45.16% <83.33%> (+0.35%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5a712db...9febe77. Read the comment docs.

[rneher]

I was initially worried that this would come at a big performance cost, but this doesn't seem to be the case.

[huddlej]

Thank you for testing this out, @rneher! I'll merge now.