Add a feature to sample a fixed number of reads

[bwlang]

Sometimes it's nice to sample a fixed number of reads from many libraries during analysis. I think it's not important that the exact number of reads be the same for every library, which should allow more efficient sampling approaches. E.g. it might be possible to do this in one pass using the file size and the first N reads to estimate the total number of reads.

Specifying a fraction of total reads is fine, but harder to use.