Add grouped parallel I/O for graph placements of unpaired reads

[Sagorikanag]

Summary

Adds parallel grouped single-read alignment iteration for multiple graph placements sharing the same read name.

This will be needed by vg surject --diploid-map, where the primary and secondary graph placements for a read must be processed together.

Changes

• Add a generic grouped_for_each_parallel iterator.
• Add grouped GAM input using ProtobufIterator.
• Add grouped GAF input by query name.
• Perform expensive GAF-to-Alignment conversion on worker threads.
• Return the number of processed read-name groups (runs).

Input requirement

Records belonging to the same read must be consecutive in the input.

[gaoj66-roche]

@adamnovak @Sagorikanag Is this PR ready? Once this is merged I can rebase my PR for rescuing/promoting secondary alignments, although in order to fully support the features introduced in my PR, we also need a grouping iterator for interleaved input.

[adamnovak]

I still have to do a full review, but I think this is on the right track.

The interleaved reads we think won't just automagically work because they're going to have different read names (/1, /2, etc.) and thus different keys, right?

But @Sagorikanag, aren't you already dealing with paired-end data in diploid surjection, using this code? How do you end up keeping the pair partners together through this?

[Sagorikanag]

Hi @adamnovak @gaoj66-roche
Yes, these wrappers are for non-interleaved reads only, I have the wrappers for the interleaved reads too, just doing some more testing as they interact with the new fragment length distribution learning I've incorporated. Will have a PR for them soon.