Add and test GR tag during surject (plus implement some kind of post-processing)

[CormacKinsella]

• surject has the --graph-aln option to preserve the graph space truth in a GR tag (supplements the regular CIGAR showing differences to the linear backbone).
• E.g. alt vs reference found in CIGAR + MD, but perfect match in GR
• likely fully ignored by downstream tools, so would require post-processing for possible interpretations
• GR tag is cs-style

[CormacKinsella]

• Check if it encodes the haplotype also, probably impracticle in many cases, e.g. traverses multiple paths but alt w/ respect to ref

[CormacKinsella]

AI suggestion on how to actually recover the aligned haplotype, not assessed this at all:

To actually recover haplotype identity you'd want:

The full GAM output from giraffe, which contains graph node traversals
The AS and XG type tags that vg produces
Keeping the pre-surjection GAM and cross-referencing with GBWT path names

[CormacKinsella]

Potential post-processing of a GR tag...

paftools.js can work on cs strings, however BAM to PAF format conversion (paftools.js sam2paf input.sam > output.paf) will produce the cs strings from the MD + CIGAR, probably doesn't retain the GR tag. So either custom conversion or another tool, e.g. pysam, samtools + unix tools with custom parsing

Meangful things that could be extracted (AI suggestion):

| Graph-perfect reads that look variant vs ref | GR tag shows all : ops but CIGAR has mismatches |
| Surjection quality degradation | Compare NM tag vs mismatch count in GR |
| Filter high-confidence alt-allele reads | GR perfect + NM > 0 |
| Identify surjection-mangled reads | Large discrepancy between GR and CIGAR |