Update to sourmash Nodegraph instead of khmer to allow for computation of minimum read length

[olgabot]

With this PR: sourmash-bio/sourmash#1009 - the sourmash Nodegraph now becomes more appealing to use instead of khmer's Nodegraph. This is because the n_unique_kmers attribute that is now added, allows for computation of the minimum necessary read length for a given false positive rate with this equation:

[olgabot]

This is low-hanging fruit for speeding up the module, but backwards compatibility may be tricky for older bloom filters. There's probably a fairly straightforward try/except thing to use here, though.