feat: support to_hgvs for single-residue protein variants

[developer-rpai]

Summary

Closes #633 — adds to_hgvs support for protein variants, scoped to single-protein-residue changes as requested in the issue.

Problem

HgvsTools._to_sequence_variant raised ValueError("Only nucleic acid variation is currently supported") for any allele on a protein sequence (e.g. an NP_ accession), so Translator.translate_to(allele, "hgvs") could not export protein alleles like the issue's example (NP_060204.1:p.Val261Ala).

Approach

Protein variants carry the reference residue in the position itself (HGVS protein notation: p.Val261Ala), unlike nucleic variants where ref/alt live in the edit. The fix adds HgvsTools._to_protein_sequence_variant, which:

• accepts only single-residue changes (end - start == 1) with a LiteralSequenceExpression state — anything else (multi-residue spans, insertions, ReferenceLengthExpression states, reference alleles, invalid/multi-residue alt codes) raises a descriptive ValueError;
• fetches the reference residue via the data proxy and builds the variant with hgvs.location.AAPosition + hgvs.edit.AASub (substitutions, incl. nonsense * → Ter) or hgvs.edit.AARefAlt (single-residue deletions, empty alt → del);
• runs the constructed expression through the existing parse + hgvs-normalize step (same HGVSDataNotAvailableError fallback as the nucleic path), so the reference residue is validated against UTA when UTA data is available.

Tests

New tests/extras/test_to_hgvs_protein.py (9 tests, all passing):

• the exact allele from Support to_hgvs for protein variants #633 → ["NP_060204.1:p.Val261Ala"]
• nonsense substitution → NP_060204.1:p.Val261Ter
• single-residue deletion → NP_060204.1:p.Val261del
• negative controls (all assert ValueError): multi-residue span, insertion (start == end), ReferenceLengthExpression state, reference allele (alt == ref), multi-residue alt, invalid residue code

The new tests use a stub data proxy and a mocked UTA connection, so they run without SeqRepo/UTA. ruff check and ruff format --check pass on both changed files.

Not tested locally

• The UTA-backed normalize() validation path for protein variants (no UTA database available in this environment); the no-UTA-data fallback is what the tests exercise.
• The full existing test suite, which requires SeqRepo data and external services not available here; no existing code paths were modified beyond delegating the sequence_type == "p" branch.