Create an iterator over `PileupColumn`s from a list of `AlignedSegment`s

[nh13]

Currently, to create an iterator over PileupColumns, we need to call AlignmentFile.pileup().

If instead we have a list of AlignedSegments and want to a pileup, we have write it to a file and return the pileup, which is pretty ugly and inefficient.

I wanted to inquire about the appetite for adding support for this into pysam via a PR.

Sketching this out quickly to prove to myself this is feasible, I was thinking the following:

1. Create a htslib pileup struct (bam_plp_t) using bam_plp_init.
2. Push all the records (bam1_t) onto the pileup using bam_plp_push.
3. Call bam_plp_next to yield the next pileups (in the next method of the iterator).
4. Use bam_mplp_t to iterator over the range of possible positions (assumes the records are in coordinate sorted order when pushed on the underlying iterator)
5. Use the above to implement the IteratorColumn interface.

If that sounds like a reasonable implementation sketch, I'd be glad to make a PR.
I made a draft PR for comments before I go any further (e.g. tests): #1353

[jmarshall]

I think this would be an excellent piece of functionality.

It would be good to take an iterable in general rather than a list specifically. I think it is indeed fair to require the input records to be sorted.

Judging by the existing functionality using these bam_plp_* functions, it looks like it should be possible to use the bam_plp64_auto() entry point — which might or might not make implementation easier. More usefully, I'm guessing it should be possible to emit columns and only read from the input iterable as required rather than reading them all up front.

[nh13]

See: #1353

[nh13]

@jmarshall I think the functionality without lazy loading of the records (i.e. "read from the input iterable as require") is useful to start. Would you be open to a second PR (after this is added) for that functionality?

#1353 is updated with docs. If you think it's the right path, I can see about adding some tests.