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/p3.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+# Phase P3: refit normalization transforms (1.2.0 and 1.1.2) for every vendor GTC and compare bit by bit.
+# Args: <data dir with bpm and data/>  <results dir>
+set -uo pipefail
+D=$1; OUT=$2
+/usr/bin/time -v .venv/bin/python scripts/validate_p3_large.py --root $D/data --bpm $D/HumanOmni2.5-4v1-Multi_B.bpm \
+  --out $OUT/summary.json --workers "$(nproc)" 2> $OUT/time.txt || echo "P3 validation reported differences"
+tail -25 $OUT/time.txt