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/trial.sh

@@ -18,6 +18,7 @@ MAX_PRICE=${MAX_PRICE:-0.30}        # USD/hour spot cap (on-demand is 0.6528)
 MAX_MINUTES=${MAX_MINUTES:-360}     # hard cap: 6 hours
 DISK_GB=${DISK_GB:-250}
 BUDGET_USD=${BUDGET_USD:-10}
+TASK=${TASK:-p3}                 # scripts/cloud/tasks/<TASK>.sh
 TAG=simurg-trial
 ACCOUNT=$(aws sts get-caller-identity --query Account --output text)
 BUCKET=simurg-trial-$ACCOUNT
@@ -60,10 +61,12 @@ launch)
   echo "-> packing repository at $(git rev-parse --short HEAD) -> s3://$BUCKET/$RUN/"
   git archive --format=tar.gz -o /tmp/simurg-repo.tar.gz HEAD
   $AWS s3 cp /tmp/simurg-repo.tar.gz "s3://$BUCKET/$RUN/repo.tar.gz" --only-show-errors
-  sed -e "s/__BUCKET__/$BUCKET/" -e "s/__RUN__/$RUN/" -e "s/__MAX_MINUTES__/$MAX_MINUTES/" \
+  [[ -f scripts/cloud/tasks/$TASK.sh ]] || { echo "unknown TASK $TASK"; exit 2; }
+  sed -e "s/__BUCKET__/$BUCKET/" -e "s/__RUN__/$RUN/" -e "s/__MAX_MINUTES__/$MAX_MINUTES/" -e "s/__TASK__/$TASK/g" \
     scripts/cloud/userdata.sh > /tmp/simurg-userdata.sh
   AMI=$($AWS ssm get-parameter --name /aws/service/canonical/ubuntu/server/24.04/stable/current/arm64/hvm/ebs-gp3/ami-id \
     --query Parameter.Value --output text)
+  echo "-> task: $TASK"
   echo "-> launching $TYPE spot (cap \$$MAX_PRICE/h, hard limit ${MAX_MINUTES} min, ${DISK_GB} GB disk deleted on exit)"
   ID=$($AWS ec2 run-instances --image-id "$AMI" --instance-type "$TYPE" --count 1 \
     --iam-instance-profile Name=$ROLE \