Reduce memory

[brentp]

This PR needs to be tested end-to-end. But, it makes small, simple changes to reduce memory in merge.

The first change does some nim stuff to reduce memory to about half without changing any of the algorithm. This gives about 2X reduction in memory. This also reduces runtime by more than 2X

The second set of changes adds a --chromosome argument to merge so that it will only read and consider reads from a given chromosome. For chromosome 1, this will give a 12X reduction in memory. With these changes, we can run 50K samples with about 256GB of memory.

[hdashnow]

I confirmed that this runs end-to-end without error on a handful of 1000 genome files and that the by-chromosome strategy produces the same bounds. Any other tests you think I should do?

I'm also going to add an example of a workflow with the by-chromosome strategy.

[brentp]

The changes are quite minimal. I think it's fine to merge.