Improve pandora discovery RAM usage

[leoisl]

I am quite concerned with this part in denovo racon: https://github.com/rmcolq/pandora/blob/12a08c5483c19fc12411e174970d31c86e842a2d/src/denovo_discovery/discover_main.cpp#L205-L206

This is a dictionary from loci names to the subreads that map to each locus, inferred by pandora map. This structure could get potentially very large, as we basically store a substring of every read that map to each locus (is just the region of the read that maps to that specific locus, but still...). There are potentially many better ways to store this info, but I also want to avoid premature optimisation, and just work on this if RAM is indeed an issue.

Originally posted by @leoisl in #303 (comment)

[iqbal-lab]

Thumbs up