Fix 6 effect-annotation bugs, add 32 regression tests

[iskandr]

Summary

Closes or substantially addresses six reported bugs plus one visibility improvement, with 32 new regression tests (378 → 410 passing, zero regressions).

Classification bugs (wrong effect type returned)

• Insertion effect from SNV? #250 — SNV in the stop codon reported as Insertion(p.XinsR) instead of StopLoss when the 3' UTR begins with another stop codon. Root cause: translate_in_frame_mutation had branches for first_utr_stop_codon_index > 0 and == -1 but not == 0, so using_three_prime_utr stayed False and the StopLoss branch was never taken.
• Stoploss confused as insert #205 — Same root cause as Insertion effect from SNV? #250 on a different transcript.
• Synonymous Insertion before stop codon annotated as non silent insert #201 — Insertion of ATATAA before the stop codon is reported as Insertion(p.1291insI) even though the resulting protein is identical to the reference (the inserted `TAA` creates a new stop at the same protein position). New silent check: if a premature stop is introduced and the new amino acids match the reference tail, classify as Silent.

Position / format bugs

• Silent.aa_pos wrong for synonymous SNPs #208 — `Silent.aa_pos` was off by one. After `trim_shared_flanking_strings`, `aa_mutation_start_offset` had been advanced past the shared prefix, but for Silent we want the position of the first affected codon. Fixed by subtracting `n_aa_shared` in the Silent call.
• Fix short_description for PrematureStop with empty aa_ref #216 — `PrematureStop.short_description` produced `p.7*` when `aa_ref` was empty (the stop was introduced by an insertion, not a substitution). Now returns HGVS-style `p.7ins*` / `p.7insK*` when `aa_ref` is empty.
• Use HGVS format for Silent.short_description #217 — `Silent.short_description` returned the literal string `"silent"`. Now returns HGVS-standard `p.R6=`.

Loader robustness

• varcode fails when it sees strange sequences in a vcf #88 — VCFs with symbolic alleles (``, ``, `INS:ME:ALU`) or breakend notation (`G]17:198982]`) caused `load_vcf` to crash. These are now skipped with a visible `warnings.warn` call that reports the total skipped count per file and points at the tracking issue (Support VCF symbolic alleles and breakends as extended variant types #264) for eventual structural variant support.

API ergonomics

• EffectCollection is not sorted by priority by default #227 — `EffectCollection` now sorts by effect priority descending by default. The first element is the most severe effect. Pass `sort_key=False` to disable sorting or your own callable to override.

Test plan

• `tests/test_stop_codon_classification_bugs.py` — 4 tests for Insertion effect from SNV? #250/Stoploss confused as insert #205/Synonymous Insertion before stop codon annotated as non silent insert #201
• `tests/test_silent_aa_pos.py` — 3 tests for Silent.aa_pos wrong for synonymous SNPs #208
• `tests/test_symbolic_alleles.py` — 7 tests for varcode fails when it sees strange sequences in a vcf #88, including a test that `warnings.warn` is emitted with the correct count
• `tests/test_premature_stop_short_description.py` — 5 tests for Fix short_description for PrematureStop with empty aa_ref #216 covering all aa_ref/aa_alt permutations
• `tests/test_silent_hgvs_description.py` — 4 tests for Use HGVS format for Silent.short_description #217
• `tests/test_effect_collection_sort_order.py` — 3 tests for EffectCollection is not sorted by priority by default #227 including verification that custom sort_key is still respected
• Existing MAF test updated since varcode's `short_description` now aligns with MAF's own `p.G384` convention rather than needing a special case for `"silent"`
• Full test suite: 410 passed, 0 failed
• `./lint.sh` passes

Related

• Spawned Support VCF symbolic alleles and breakends as extended variant types #264 (tracking issue) for eventual first-class support of symbolic alleles and breakends as extended variant types (depends on Add structural variant types (translocations, inversions, duplications, breakpoints) #257).