task_counts.sh: featureCounts should use the unsorted BAM/CRAM

[gpertea]

Weirdly enough, featureCounts requires input BAM/CRAM to be sorted by read name instead of coordinate (in order to keep paired reads together).
If coordinate-sorted input is given, featureCounts first sorts the input stream by name using huge temporary files (because they are uncompressed SAM!).

Since task_fq2cram.sh should be keeping the "unsorted" BAM/CRAM file around, that one should be served directly to featureCounts in this mini-RNAseq pipeline.