P2 (WIP): raw X/Y assembly, canonical Parquet model, simurg ingest

- signal/raw_xy.py: vectorised Infinium I/II raw X/Y and control assembly
  (plan §7.1) with a per-chip address-map cache; IN-027 coverage gate
- core/schemas.py: Arrow schemas (loci, sample_intensity, controls),
  deterministic atomic Parquet writing
- pipeline/ingest.py + CLI: discovery, sheet, IN-010 naming, overwrite
  check before any work, spawn-based workers, per-sample isolation,
  --resume by input hashes, run manifest, exit codes 0/3/4
- matches gtc2vcf raw X/Y and controls exactly on 18 GSA samples;
  outputs byte-identical across thread counts and layouts

pyproject.toml

@@ -15,6 +15,7 @@ dependencies = [
     "typer>=0.12",   # CLI
     "rich>=13",      # progress / human-readable output
     "pyyaml>=6",     # --config files
+    "pyarrow>=15",   # canonical columnar model and Parquet outputs (plan §5)
 ]
 
 [project.optional-dependencies]