Change raw_combine_files to local rule

[tanaes]

All this does is cat fastqs together; submitting as a separate job to scheduler probably is actually adding overhead to cluster.