Verify real Sid RNA phasing through protein effects

[iskandr]

Real Sid RNA phasing is now tested through VariantCollection.effects(), so
the tests verify the protein produced by a phased combination, as well as the
read evidence establishing that combination. NTF3's two individually predicted
p.K56R / p.K56N edits yield AAG>AGT, p.K56S and the independently expected
DVSENY peptide when the real reads establish cis. Real trans evidence and
insufficient support retain separate DNA predictions and no joint prediction.

Extend this to MAP2's two substitutions plus a 28-base deletion using the
existing anchored-sequence evidence API. Original T1/T2 bulk RNA witnesses with
28N gaps support the same local compound sequence as the single T2 ONT witness
with 22D2M6D; only the bulk evidence clears the unchanged two-template floor.
The combined 904-aa reference-frame protein is checked independently. Corrected
complex, deletion-only and historical 22-base alleles remain distinct, with
their DNA predictions retained even when their RNA hypotheses do not match.

• Keep every available original record of selected RG/QNAME templates in the
  pinned neutral openvax-v2 bundle; verify record digest multiplicities against
  the complete source subset, including mates and supplementary placements.
• Add an independent 16.8 KB reference asset acquired from versioned NCBI
  RefSeq mRNA/protein and GRCh38 genomic records. Pin source receipts, CDS edits,
  codons, full protein/CDS hashes and stops; no Varcode/PyEnsembl imports were
  used to derive the expected effects.
• Exercise NTF3 with fast, protein_diff and transcript_model, and MAP2 with the
  established fast/protein_diff annotators. Verify individual effects across
  transcripts and joint membership/protein/phase provenance after JSON.
• Document reference-frame/local-RNA limits and bump the patch version to
  10.11.6. This adds integration coverage and references; runtime semantics,
  dependencies and global phasing thresholds are unchanged.

Validation: ./lint.sh passes; 99 focused phasing/haplotype/packaging tests
pass; full TEST_SH_MAX=2 ./test.sh passes with 3,234 tests (22 existing
warnings, 219.34s). Fixture inventory and source-distribution checks pass.

Refs #464. The audit reproduced existing defects filed separately as #574
(experimental effect JSON drops realized candidates/hypotheses/products) and
#575 (RNA support cache uses the final fragment key). Experimental complete-CDS
checks use in-memory products; JSON checks assert its currently preserved
protein, membership and phase source. Returned support counts are correct in
these fixtures. This PR does not claim completion of the broader #464 panel.