Native methylation-aware phasing: wire the ported algorithm into make_examples

[nh13]

Stacked on #7 (which is stacked on #5) — review after they merge. Opt-in (--enable_methylation_aware_phasing, default off); default behavior is byte-identical. Apple-only native code; nothing ports upstream.

Depends on #7 for its prerequisites (aux-field parsing + the single-sample multi-sample VariantCaller). This PR is now just the methylation-aware phasing feature itself.

What this does

Wires the already-ported, unit-tested methylation_aware_phasing algorithm into the make_examples single-sample phasing block (mirroring make_examples_core.py): methylated reference sites (alt='.') are split out of the SNP phasing graph, DirectPhasing runs on the SNP candidates, and PerformMethylationAwarePhasing then assigns a haplotype (Wilcoxon rank-sum vote on 5mC) to reads DirectPhasing left unphased. The refined per-read phases feed read_hp_tags / the haplotype small_model exactly as the DirectPhasing output does.

• The option is set on AlleleCounterOptions/VariantCallerOptions/MakeExamplesOptions.
• dv_methylation_aware_phasing is linked into dv_make_examples_lib.
• A log line reports methylated-ref-site count + reads rephased when the path engages.

Scope: single-sample path (the primary PacBio/ONT methylation use case); per-site p-values are dropped (upstream uses them only for --phasing_error_stats_output). Like --enable_methylation_calling, the flag is exposed at the make_examples stage, not forwarded through run.

Validation

• 14/14 ctest green (incl. the upstream methylation_aware_phasing unit test run against the native build). Flag off (default): byte-identical.
• Real PacBio HiFi 5mC data (PacBio's public HG002-CpG-methylation-202202 BAM, chr20 ~40×, streamed via a .bai-guided byte-range fetch): with Native single-sample correctness: multi-sample VariantCaller + aux-field parsing fix #7's prerequisites in place, the active path engages — 6 methylated reference sites per region, PerformMethylationAwarePhasing runs on real 5mC. (At 40× HiFi DirectPhasing already phases all reads, so the vote correctly finds 0 additional reads to rephase; the algorithm executing on real methylation data is the end-to-end proof. A sparser-SNP locus would exercise the rephase-count path.)
  • Note: the public dataset uses legacy lowercase Mm/Ml tags (pre-2021 SAM convention); DeepVariant expects uppercase MM/ML. Renaming them (semantically identical) is required for that specific file — modern PacBio/ONT outputs use the uppercase tags.

[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.