No longer merge fastq files from different read groups together

No longer merge the raw fastq files, but trim and map them per readgroup. The different read groups are merged in the MarkDuplicates step, which accepts multiple bam files.

This is faster because it prevents an additional read/write of all the raw fastq data, and because more of the pipeline can be run in parallel when there are multiple readgroups.

[] Try to fix gitlab-ci in this branch