fix(variants): read UniProt-native deletions (empty <variation>)

[trishorts]

What

Fixes the ProteinXmlEntry reader so it reads UniProt-native deletions, which UniProt encodes as an empty <variation/> (e.g. <original>P</original><variation/>). The reader previously admitted a sequence variant feature only when <variation> was non-empty, so every UniProt-native deletion was silently dropped — no SequenceVariation was ever created. This is not something we can push onto the encoding: it is how UniProt writes a deletion.

The guard now admits any feature that encodes a real change (not both <original> and <variation> empty), so deletions are read alongside substitutions and insertions. SequenceVariation, the writer, and the round-trip already support an empty VariantSequence, so no other production change is needed.

Corpus (L1 deletions — executable spec, oracle derived from the domain model, red-before-fix)

New nodes in Test/DatabaseTests/VariantCorpus:

• D00–D01 — single- and multi-base deletion, unmodified
• D02–D04 — deletion before / on / after a PTM residue (phospho re-anchors ↓ / drops / stays)
• D05–D06 — deletion out of range / straddling the C-terminus → pruned
• D07–D08 — the VCF depth cutoff (minAlleleDepth), same deletion passing at cutoff 5 (2 forms) / filtered at 7 (1 form)
• MD00–MD02 — pairs of separate deletions (unmodified, flanking a PTM, one-on-the-PTM), capped at 2 to stay under the combination cap

D00–D04 fail on master (reproducing the reader gap) and go green with the fix. The harness gains BEGIN/END, empty-VAR (deletion), SRC=vcf (GT/AD/DP) support, and a per-case MinAlleleDepth.

Buggy fixture the drop was masking

Admitting the deletions surfaced a latent error in seqvartests.xml: the YPSE deletion declared begin=960, but YPSE sits at 961–964 (the base is NYPSE at 960–964), violating SeqVarXmlTest's own invariant that OriginalSequence equals the residues at [begin,end]. It never ran because the variant was being dropped. Corrected begin 960→961.

Verification

• 34/34 VariantCorpus (incl. ProteinDbWriter round-trip + run-twice determinism)
• 666/666 DatabaseTests + Transcriptomics

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.46%. Comparing base (77fb7db) to head (ca57af0).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1095   +/-   ##
=======================================
  Coverage   91.46%   91.46%           
=======================================
  Files         427      427           
  Lines       52777    52777           
  Branches     6325     6325           
=======================================
  Hits        48273    48273           
  Misses       3268     3268           
  Partials     1236     1236           

Files with missing lines	Coverage Δ
mzLib/UsefulProteomicsDatabases/ProteinXmlEntry.cs	98.94% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[trishorts]

Manual unit-test review

All 14 unit test(s) added by this PR were reviewed individually by the author (not only by automated/AI-generated checks): 13 approved as written, 1 edited, 0 flagged.

This satisfies the author's standing AI-use policy that AI-assisted tests receive human review before relying on them. Recorded 2026-07-11 06:54.