Create new "GFA tabix" adapter for multi-way synteny visualizations

[cmdcolin]

Background

JBrowse 2 can show multiple pairwise comparisons but is is not very smooth
a) you have to manually go through several steps in the linear synteny import form
b) it uses multiple files e.g. human vs mouse paf, and mouse vs rat paf
c) it also takes up a lot of visual space on the users screen: each row is a full featured linear genome view, and cannot be collapsed down to a single 'line' currently

This PR

This PR designs a new data format called "GFA tabix" to help index into a pangenome graph (containing multiple whole genome assemblies) and adds several custom visualization to help incorporate it into jbrowse 2

• Uses a adapter called GfaTabixAdapter
• This adapter is able to actually query from any different species to any other species efficiently (graphs are effectively ALL vs ALL, and our tabix encoding preserves this)
• We created a MultiSyntenyTrack type, with a MultiLGVSyntenyDisplay (analogous to LGVSyntenyDisplay for pairwise alignments) for showing each sample in the graph as a row with alignment vs the reference on a normal linear genome view
• Users can then, for a region of interest, launch a full featured linear synteny view, or even graph genome view (e.g. our js version of bandage plot) when they are viewing their MultiSyntenyTracks

Screenshots

4-way arabidopsis from single file

Volvox_del.fa and volvox_ins.fa vs the reference volvox (simple demo data)

HPRC mitochondrial genome 44 samples

Data format

We made a rust program that iterates over the converter walks every path
once, accumulating segment lengths to compute genomic coordinates, then
creates two files

Firstly, it creates a tabix-indexed file called pos.bed.gz that maps genome coordinates (e.g. you are browsing region of interest on hg38) to the GFA data that overlaps that region, and gives you the graph 'segment ids'

Then you can secondarily query another file: segments.gz (a bgzip compressed + gzi indexed+ segment id-> byte offset indexed, kind of like fai for fasta) that we created that actually does the reverse, and maps each segment back to the genome coordinates for genomes that pass through that segment of the graph.

Crucially, pos.bed.gz does not have "reference bias" because it encodes the entire graph and can be queried from any genome in any way you want. You query it using PanSN naming scheme which is just a simple way to prepend assembly name essentially to the chromosome name, and thus stores data from every assembly in that single file as a result https://github.com/pangenome/PanSN-spec

That is the gist of it. I was very happy to make some progress on it so just making a very early draft PR as usual

Note: I also looked at existing tools that are sort of similar, and https://jmonlong.github.io/manu-vggafannot/ ('gafannot') was a helpful motivator. It has some similar concepts but gafannot indexes something called GAF (graph alignment format) e.g. high throughput reads or even gff3 features aligned to the graph, whereas we are indexing the graph (GFA) itself!

Note: i used claude code extensively to draft this and it will need further review and validation. A more 'ai generated summary' is in PR_SUMMARY_SIMPLIFIED.md in this branch but i didn't want to subject people to raw AI generated text 🤖