Apply seqtk PR to improve kseq.h parsing performance

[jmarshall]

This applies @kloetzl's lh3/seqtk#123 to klib's kseq.h.

For one large data file on my machine, this reduces _kseqbench2's kstream time from ~1.65s to ~0.92s. (I have not heavily tested it otherwise.)

[attractivechaos]

Thanks!