v1.3.0: gnomAD Exome Frequency Cache

Added

• gnomAD population allele frequencies. New GnomadAnnotator enriches
  report annotations with population frequency context from gnomAD v4.1 exomes
  (~16M variants, 730K individuals). Pre-built cache downloaded from HuggingFace
  via db update. Frequency column in terminal, HTML, and JSON reports.
  --no-gnomad flag to skip.
• CPIC fallback for PharmGKB. db update succeeds when CPIC API is
  unreachable — reuses cached allele function data. Recovery auto-triggers on
  next successful check.
• Graceful db update. Individual annotator download failures print an error
  and continue to remaining annotators instead of aborting the entire update.
• scripts/build_gnomad_cache.py — streaming VCF build script. Downloads ~120GB
  gnomAD exome VCFs over HTTPS (or reads local files with --local-dir), never
  saves VCFs to disk, outputs ~6GB SQLite (~3GB gzipped).
• JSON report schema_version bumped to "2" (added allele_frequency field).
  Diff engine accepts both v1 and v2 baselines.
• gnomAD ODbL v1.0 attribution in HTML and JSON reports.
• CI workflow (.github/workflows/ci.yml) — lint + test on push/PR to main.

Fixed

• Offline claim in README corrected: analysis runs offline by default with opt-out
  freshness check, not opt-in network access.
• __del__ partial-init crash on PharmGKB constructor failure.
• .gitignore updated for GWAS Catalog test data.

Changed

• Pre-push hook reduced to version-tag check only (CI runs the full suite).

Technical

• Composite primary key (chrom, pos, ref, alt) on gnomad_frequencies —
  preserves multi-allelic sites (rsID-only PK silently dropped ~20% of records).
• Coordinate columns indexed for future AlphaMissense/CADD integration.
• MAX(af) GROUP BY rsid in lookup queries handles multiple rows per rsID.
• 951 tests, 93%+ coverage.