Validate real genomes: phase sets, canonical QUAL digest, batched shapes

[ecrum19]

On a real genome (NG131FQA1I, first 250,000 records, 58.2M triples), v3.3.0's validation had two problems:

• it reported a mismatch, even though every field of every record equals the VCF text;
• its shape checks exhausted a 31 GB host.

None of these were conversion errors; all three causes were in the validator. This PR fixes all three.

Problem	Fix
Q9/Q10 mismatch: the oracle didn't count GATK's phase sets (30,910 vcfc:PhaseSet)	The census oracle now counts what the emitters write for phase sets, SV events, confidence intervals (including FALDO) and gVCF reference blocks. test_validation_real_files_unit.py checks the oracle against both emitters, for VCFv4.2 and 4.5.
Q11 mismatch: QLever prints a decimal QUAL canonically (30.10 → 30.1), so 788 digest buckets differed	Q11 and the oracle both drop a decimal QUAL's trailing zeros before hashing. That changes 1,998 values, and 0 buckets then differ.
pyshacl loaded the whole graph	Node-local shapes (the default core profile) are now checked a batch of whole records at a time, in parallel. Each batch also carries the file-level triples it points at, so the verdict is the same as checking the whole graph.

New options

• --shacl-batch-triples (default 500,000) and --shacl-workers (default up to 4).
• Measured with the core shapes: about 1.3 GB per batch. Time grows in step with batch size, so smaller batches cost no throughput.

Size limit for checking the whole graph

• Shapes with SPARQL constraints (the full profile) are still checked as one graph, so they keep a size limit, --shacl-max-triples.
• That limit drops from 50M to 10M triples. At the measured ~2.4 GB per million triples, 50M would need about 115 GB. The old comment said the campaign had shape-checked 17.1M triples, but its largest was 0.96M.
• The wrapper no longer limits the core profile by size, since it is batched now.

Verification, all on bench-2 inside the rebuilt image:

• Full suite: 1,284 tests OK (60 skipped).
• End to end on NG131FQA1I's slice: validation PASS.
  • 13/13 comparisons equal.
  • Shapes PASS: 117 batches, 0 violations, 46 warnings. The warnings are VCF 4.5 recommending Source and Version on INFO declarations.
  • The shape checks took 89 min, and available memory never fell below 19.9 GB.
  • The earlier attempt with 2M-triple batches was stopped by the watchdog when available memory fell to 0.6 GB.

Docs updated: validation.md, validation-methodology.md, validation-migration-notes.md, cli-reference.md.

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 91.04803% with 41 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
test/test_validation_real_files_unit.py	88.63%	25 Missing ⚠️
src/validation/validation_runner.py	91.11%	16 Missing ⚠️

📢 Thoughts on this report? Let us know!