Speed up bam splitting for bcall

[AntonJMLarsson]

I made two changes to split_barcoded_bam.py which speeds it up roughly 10-fold:

• Uses has_tag and get_tag to fetch the barcode per read, instead of iterating over all tags.
• Constructs a dictionary to match the read barcode to the whitelist (dictionary keys) and fetches the correct file by using that dictionary, instead of checking if the barcode is in a list and then getting the list index value for that barcode.

Both changes turn these two tasks from linear-time to constant-time operations, thereby speeding up the overall code.

Best, Anton

[caleblareau]

This looks great! Thank you for the PR!! Can I ask about the 5 threads in the Alignment file import? should I parameterize that or is that a magic number?

[AntonJMLarsson]

I've found that it usually helps a little bit to speed up the reading. More than 5 threads usually have diminishing returns, so having more than that is probably not helpful. But as you can see I decided to remove it in the next commit, since it should depend on the -c option and that is not currently an argument to this particular script. If it was implemented as an argument I'd do something like threads = min(5,cores) to ensure not more cores are being used than what has been specified.

Regardless, the main speed-up is certainly the two main changes detailed in the first comment.