usability improvements

As of this version gargs will not read everything into memory as before. It will read up to 1MB. If it does not get an EOF by then it start using a temp-file. This will reduce memory usage.

It also fixes --nlines to be quite useful e.g.:

cat regions.txt | gargs -p 20 -n 10 "bcftools view some.bam {}"

will send 10 regions to each process to amortize the cost of loading the index into memory. Note that the place-holder {} is specified only once.

It also defaults --sep to "\s+" if --nlines is not specified.

Finally, it adds a --retry argument that takes an integer that indicates the number of times a failed process should be retried. This is nice for transient network errors.