perf: reduce memory footprint of cyrcular graph breakends command

[dlaehnemann]

• reduce max usable read coverage to 32767 by switching Count to i16 from i32, this just about halves memory usage

[dlaehnemann]

Also, in a second step, could we go and simply use mosdepth for the depth calculation and have it output a D4 file. From what I understand writing these in mosdepth is really fast; and they can be read with random access and have a whole rust crate built for them](https://docs.rs/d4/latest/d4/).

[tedil]

Also, in a second step, could we go and simply use mosdepth for the depth calculation and have it output a D4 file. From what I understand writing these in mosdepth is really fast; and they can be read with random access and have a whole rust crate built for them](https://docs.rs/d4/latest/d4/).

Yes, I don't see why not (apart from requiring additional external input files)

[dlaehnemann]

Yeah, but it would also be more in the spirit of specialized tools that do one thing, but do it well... But I have to admit that the D4 crate does not come with any examples in the docs, so one would probably have to scour the d4tools implementation to get the random access file reading set up.

[dlaehnemann]

Also, this PR here seems to have fit all of my current data into the respective HPCs memory. Just barely, but it worked. So I am assuming there's probably also another code path that uses a lot of memory at some point. I'll have to run heaptrack with a large sample on the HPC at some point.