Support GAM subsampling to mapeval on large read counts.

[glennhickey]

GAM compare logic isn't going to scale to, say, whole-genome 30x coverage. Resolving #471 would likely help, but the lack of BAM support in vg gam compare would still be a bottleneck. One work-around is subsampling the GAMs (new vg filter option?) and working on those.

[adamnovak]

I have an implementation of this in my "wholegenome" branch of toil-vg; I added a new option to vg filter to do the subsampling. I'm going to wait until I have a vg Docker I like that supports it, to plug into the toil-vg config, before I PR it.