Fix crash displaying modifications called on softclipped regions of reads

[cmdcolin]

Found a crash displaying modifications, example share link
https://jbrowse.org/code/jb2/v1.7.0/?config=test_data%2Fconfig_demo.json&session=share-Jfg7Adjvai&password=ZCjW1

This is probably due to displaying a modifications that was called on a soft clipped area of a read (e.g. a part of the read that did not properly map to the reference genome)

Instead of trying to display these (and currently, causing a crash) this PR proposes to filter them out

[codecov]

Codecov Report

Merging #2964 (c6f9b84) into main (8b9d869) will increase coverage by 0.02%.
The diff coverage is 72.05%.

@@            Coverage Diff             @@
##             main    #2964      +/-   ##
==========================================
+ Coverage   60.67%   60.69%   +0.02%     
==========================================
  Files         601      601              
  Lines       27342    27361      +19     
  Branches     6654     6665      +11     
==========================================
+ Hits        16589    16606      +17     
- Misses      10442    10444       +2     
  Partials      311      311              

Impacted Files	Coverage Δ
...lugins/alignments/src/LinearPileupDisplay/model.ts	64.97% <ø> (ø)
...s/alignments/src/PileupRenderer/PileupRenderer.tsx	47.91% <0.00%> (-0.29%)	⬇️
...ments/src/SNPCoverageAdapter/SNPCoverageAdapter.ts	54.90% <0.00%> (-1.48%)	⬇️
plugins/alignments/src/CramAdapter/CramAdapter.ts	83.94% <33.33%> (-0.62%)	⬇️
...lugins/alignments/src/BamAdapter/MismatchParser.ts	89.09% <91.66%> (-0.02%)	⬇️
plugins/alignments/src/BamAdapter/BamAdapter.ts	72.97% <100.00%> (ø)
...ments/src/LinearSNPCoverageDisplay/models/model.ts	63.15% <100.00%> (+0.39%)	⬆️
...ignments/src/PileupRenderer/PileupLayoutSession.ts	100.00% <100.00%> (ø)
products/jbrowse-web/src/util.ts	27.27% <0.00%> (ø)
packages/core/util/layouts/GranularRectLayout.ts	87.28% <0.00%> (+0.42%)	⬆️
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c6b42a7...c6f9b84. Read the comment docs.

[cmdcolin]

This now, instead of filtering out negative coordinates in the softclipped bases, actually modifies the "getNextRefPos" to only return valid positions on the reference within the read, and so does not return areas in the softclipped bases.

This has the additional benefit of fixing a bug in getNextRefPos. This bug caused getNextRefPos to plot the wrong positions with large insertions from what I could tell. The new logic is simpler and uses some lessons learned from plotting reads without MD