P3: vectorised nearest-neighbour search; task-driven cloud trials

- fit.py: exact and binned nearest-neighbour searches use one distance
  matrix per searched bin (first-minimum rule kept); fit 2.6 s -> 1.4 s
  per GSA sample, still bit-identical to the oracle
- scripts/validate_p3_large.py: refits 1.2.0 and 1.1.2 transforms for
  every vendor GTC and reports ULP distances per field
- cloud trial runs scripts/cloud/tasks/<TASK>.sh (TASK=p3 default;
  p1p2 = previous validation + ingest)

scripts/cloud/tasks/p1p2.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+# Phase P1/P2 validation on the 1000 Genomes Omni2.5 set + `simurg ingest` at scale.
+# Args: <data dir with bpm/egt/csv and data/>  <results dir>
+set -uo pipefail
+D=$1; OUT=$2
+.venv/bin/python scripts/validate_p1_large.py --root $D/data --bpm $D/HumanOmni2.5-4v1-Multi_B.bpm \
+  --egt $D/HumanOmni2.5-4v1-Multi_B.egt --csv $D/HumanOmni2.5-4v1_B.csv --out $OUT/summary.json \
+  --workers "$(nproc)" || echo "validation reported differences (see summary.json)"
+/usr/bin/time -v .venv/bin/simurg ingest --bpm $D/HumanOmni2.5-4v1-Multi_B.bpm --idats $D/data \
+  --out /work/ingest --threads "$(nproc)" --quiet 2> $OUT/ingest_time.txt || echo "ingest exit code $?"
+cp /work/ingest/ingest_summary.tsv $OUT/ingest_summary.tsv 2>/dev/null || true
+cp /work/ingest/run_manifest.json $OUT/ingest_run_manifest.json 2>/dev/null || true