fix: align benchmarks with rebuilt :dev (noodles) RustQC image

[ewels]

Warning

This PR was generated by an automated agent (Claude Code) and has not been reviewed by a human yet. Please review carefully before merging.

Why

Issue seqeralabs/RustQC#113 (switch rust-htslib → noodles) merged on 2026-06-16, and ghcr.io/seqeralabs/rustqc:dev was rebuilt from main HEAD minutes later (now rustqc 0.2.1, commit 2556c9d, no libhts linkage — confirmed pure-Rust I/O).

That newer build changed two CLI surfaces, which broke every RUSTQC_RNA run repo-wide — this is the current red CI (ERROR ~ mapping values are not allowed here across all conda/docker/singularity shards):

1. --stranded now takes string values (unstranded/forward/reverse) instead of numeric 0/1/2 → invalid value '0' for '--stranded'.
2. rustqc --version now prints a second line (Binary: <arch> | CPU: ...), so the old sed 's/rustqc //' wrote multi-line, colon-containing text into versions.yml → invalid YAML → every versions-parsing test fails.

What changed

• modules/local/rustqc_rna.nf + rustqc_rna_profile.nf: pass string strandedness; extract only the semver for versions.yml (… | head -n1 | cut -d' ' -f2).
• Regenerated 4 stale regression snapshots against the current build (baselines were frozen at the March v0.1.x build):

  snapshot	nature of change	verdict
  dupradar	trailing whitespace removed; numeric values identical	cosmetic
  featurecounts	biotype counts now match upstream (protein_coding 39736→42331); old baseline disagreed with upstream	correctness ✅
  tin	per-transcript value drift; identical 986-gene set; re-run confirmed deterministic	benign build drift
  junction_saturation	one value max(3,3,0)→max(2,3,0); deterministic on re-run	benign build drift

• README: added samtools + tin to the comparison table; documented that :dev tracks main incl. noodles; noted RustQC no longer needs htslib/cmake to build (only g++ + libfontconfig1-dev, MSRV 1.89); added samtools to the layout + snapshot-update docs.

Validation (local, Docker, small dataset)

All 16 crosscheck-vs-upstream tests pass without snapshot changes (the task's required outcome — RustQC still matches every upstream tool). Full local results:

• ✅ upstream: 16/16 · ✅ crosscheck: 16/16 · ✅ regression: 15/15 (11 unchanged + 4 regenerated) · ✅ pipeline smoke: 4/4

nf-test work dirs were routed to an external drive; no committed config changes for that.

Out of scope / pre-existing (NOT introduced here)

• nf-core template lint (28 failures: missing .github/CONTRIBUTING.md, email_template.html, LICENSE/workflow mismatches, JSON syntax in adaptivecard.json/slackreport.json) — pre-existing template drift, unrelated to noodles.
• tests/default.nf.test has no committed snapshot on main and CI runs --ci (fails on missing snapshot). Its baseline is environment-sensitive (Linux-generated, PDF plots excluded) and is being addressed on the bigwig branch (add/rustqc-bigwig-nf-test); generating it on macOS/ARM here would itself break CI, so it is intentionally left out. The versions.yml fix here does clear the mapping values error for it.
• No versioned RustQC release tag exists for noodles yet, so --rustqc_image stays at :dev (documented).