Sample fragmented haplotypes

[jltsiren]

Changelog Entry

To be copied to the draft changelog by merger:

• Haplotype sampling can sample fragmented haplotypes in large snarls.

Description

Until now, haplotype sampling only considered haplotype paths that cross the entire block/subchain. HPRC v2 graphs have some large snarls where very few haplotypes do that. Most of them have been clipped into multiple fragments due to >10 kbp unaligned sequence intervals.

With this PR, we can sample any minimal end-to-end haplotype visit to a subchain, regardless of the number of fragments. We consider haplotype sequences consisting of a chain of paths / haplotype fragments, each of them sharing (sample name, haplotype number, sequence name). Fragments within a chain are ordered by starting offset / fragment number (fourth field in GBWT metadata). Note that due to HPRC naming conventions, we cannot cross assembly gaps this way. Because path names use accession numbers rather than chromosome names as sequence names, we cannot tell whether two assembly contigs correspond to the same chromosome and in which order.

Haplotype information (.hapl) files must be rebuilt to include fragmented haplotypes. Old files remain usable, but new files with fragmented haplotypes cannot be used with earlier versions of vg.

Also fixes #4517. GBWTGraph GFA parsing now accepts missing SeqStart/SeqEnd fields in W-lines.

I'm not going to merge this immediately, as we still need to determine the impact on read mapping and variant calling.