fix(argo-train): apply train-template.yaml before single-job submission

The multi-seed path already does `kubectl apply` before `argo submit`,
so cluster template stays in sync with source. The single-job path used
`argo submit --from=wftmpl/train` directly, expecting the cluster's
template to already match — which silently drifts when defaults change.

Caused workflow train-jpxvn (2026-05-10) to dispatch with stale
imbalance-bar-threshold=0.5 default (the cluster's old value) when the
source had been bumped to 20.0. Triggered near-OOM in feature extraction.

One-line fix: apply the template before submission. Mirrors what
multi-seed already does. No behavior change for users who pass explicit
flags; just makes implicit defaults track source.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

scripts/argo-train.sh

@@ -152,6 +152,14 @@ if [[ "$MULTI_SEED" -gt 1 || "$FOLDS" -gt 1 || "$PROFILE" == "true" ]]; then
 fi
 
 if [[ "$USE_MULTI_SEED" == "false" ]]; then
+  # Apply the local train-template.yaml to the cluster BEFORE submission.
+  # Without this, `argo submit --from=wftmpl/train` runs against whatever
+  # was last applied — defaults can drift from source. Caused workflow
+  # train-jpxvn (2026-05-10) to dispatch with a stale threshold=0.5
+  # default and trigger near-OOM. Multi-seed path already does this; the
+  # single-job path was missing it.
+  kubectl apply -n foxhunt -f infra/k8s/argo/train-template.yaml >/dev/null
+
   CMD="argo submit -n foxhunt --from=wftmpl/train"
   CMD="$CMD -p commit-sha=$SHA"
   CMD="$CMD -p git-branch=$BRANCH"