Compose control counttables for kevlar novel step

[standage]

After counting k-mers for each control sample, we should investigate composing the counttables into a single nodetable before running kevlar novel. This should a couple of synergistic benefits.

• We have a single table instead of 2 (or more), reducing time due to k-mer abundance queries
• A nodetable consumes 1/8 of the size of a counttable with the same number of buckets

The cost is, of course, another pass over the "data". But it should be possible to build a nodetable directly from the underlying counttables themselves without iterating over the reads again. So "data" should be quite small and manageable.

[standage]

Most of this would be implemented in khmer-land. See dib-lab/khmer#1379 and dib-lab/khmer#1392 for relevant threads in that project.

[standage]

Investigating this in dib-lab/khmer#1874.