Validate performance when kmer mode is supplied unpaired read data.

[cerebis]

qc3C should handle this, but performance testing is lacking. It would be nice to know how well qc3C performs in this scenario. If it is ok, we can let to happen or if we know it will be adversely affected, we can warn the user.

A previously existing warning simply checked observational redundancy (n_reads / n_pairs), and would erroneously warn users with merged short-insert libraries.