Starseqr

* Fixed a bug where fastq input was marked required but would break when run using existing alignments
* Added cwl for starseqr

* With a compendia of training data tweaked thresholds to keep FP from exploding in especially noisy samples such as FFPE.
* Added more info columns to the output

* Added bioconda recipe
* Updated basespace script
* Modified threshold for num of jxns to minfrag20 ratio from 10% to 1%. This filter was removing TPs in some datasets and other filters are catching FPs in our training datasets.

* Creates a central folder "chimeric_transcripts" for all fasta files.
* Functions now pull fastas from central chimeric transcript folder
* Reference transcripts are now extracted directly from the GTF. The (-x) parameter is no longer in use.
* Multimapping homologous fusions are considered through a network graph approach where homology and expression are used to prune clusters. If homologous the highest expressing homolog passes the filter. If they share a breakpoint but are not considered homologoues, then all partners pass the filter.
* The highest expressing fusion transcript is now part of the output
* Primers are now generated from the highest expressing transcript.

* Updates to STAR functions to use the new chimMainSegmentMultNmax parameter.
* Allow more ways to run STAR-SEQR from different STAR outputs.
* Salmon is now used to quantify candidate fusions and partner transcripts.
* Additional parameter (-x) to pass in a transcript reference for more stable counts.
* Greater sensitivity. Now rescues reads with 1 read support as long as all other filters and expression metrics are met.
* Uses more filtering approaches including basequalities, expression of transcripts relative to fusion,
* Fixed a bug so fusion junctions are now normalized to 0-base genomic coordinates.

* Fixed install script and now use sdist and twine for Pypi uploads.