Fail more obviously when SlideSort fails

Hi Martin, this keeps on biting me from time to time: when the FASTQ reads are too short (this happens when the sequencer had some automatic adapter trimming turned on), then SlideSort fails more or less silently as it is part of a pipe command. The reason why its standard error is redirected to /dev/null is that it is verbose even when it works. In this pull request, I use Bash's pipefail option to detect real errors, and make sure cagescan-find-molecules.sh exits with an appropriate return code.