Native single-sample correctness: multi-sample VariantCaller + aux-field parsing fix

[nh13]

Stacked on #5 — review after it merges. Two general correctness fixes to the native single-sample make_examples path, found while validating methylation but valuable on their own (they affect all single-sample WGS/WES runs). Apple-only; both realign the port to upstream — nothing ports upstream. Default behavior change is limited to a small, concordance-validated accuracy improvement (item 2).

1. --parse_sam_aux_fields was a no-op

It was set on MakeExamplesOptions but never on the SamReaderOptions used to open the BAM, and sam_reader.cc's ParseAuxFields does nothing unless aux_field_handling == PARSE_ALL_AUX_FIELDS. So no aux fields (MM/ML base modifications, HP, …) were ever parsed into the Read proto — read.base_modifications() was always empty, and the BAM HP tags the phasing fallback reads were never populated. Now wired (gated on the flag; default off = byte-identical).

2. Single-sample path used the wrong VariantCaller

The single-sample path constructed vcf_candidate_importer::VariantCaller (variant_calling.cc). Upstream always routes normal calling through the multi-sample VerySensitiveCaller (variant_calling_multisample.cc) for every sample count — vcf_candidate_importer is reserved for the special --variant_caller=vcf_candidate_importer import mode. Switched the single-sample path to multi_sample::VariantCaller (matching the trio path and upstream), calling CallsFromAlleleCounts with a single-element {sample_name -> AlleleCounter*} map and role sample.

This aligns the single-sample candidate set with upstream's and is a net accuracy improvement.

Validation

• 14/14 ctest green.
• Full chr20 HG002 WGS (b37) vs GIAB v4.2.1: F1 0.9974 → 0.9975, recall 0.9955 → 0.9957, precision 0.9993 unchanged (57 FP). The broader (upstream-matching) candidate set recovers a few true positives the importer caller missed, with no precision cost.

Why this PR exists

These two fixes are prerequisites for native methylation-aware phasing (PR #6, stacked on this one): without them, read.base_modifications() is empty and the single-sample path never emits methylated reference sites, so the 5mC vote can never engage. But the caller realignment changes all single-sample runs (slightly better accuracy, more upstream-faithful), so it is split out here for review on its own merits rather than buried in an opt-in feature PR.

[BenjaminDEMAILLE]

Heads-up @nh13: I had to squash apple-silicon-native-v2-pr from 273 commits down to 1, because Google's CLA bot can't auto-scan PRs with more than 250 commits and this branch is the head of the upstream PR google#1085. The squash is content-identical (same tree hash 810ff855), only the history was flattened.

Side effect: this rewrote the base of your 4 open PRs (#3, #4, #6, #7), so they now show an inflated diff against r1.10. A rebase onto the new base fixes it cleanly (no real conflicts, the content is unchanged). The old base tip was 0cd41f859de445f5d8a60524b86b54bb418a043b, so:

git fetch origin
git rebase --onto origin/apple-silicon-native-v2-pr 0cd41f859de445f5d8a60524b86b54bb418a043b <your-branch>
git push --force-with-lease

Sorry for the churn, and thanks again for all the fixes. Your authorship is preserved via a Co-Authored-By trailer on the squashed commit.