fix(argo): cap RAYON_NUM_THREADS=8 in fxcache to prevent OOM

28 parallel alpha-feature chunks × ~3GB each = 84GB peak at merge.
Capping at 8 threads keeps peak at ~24GB (well within 96Gi limit)
while still saturating I/O. This is a one-time cost — after the
cache is populated on the PVC, subsequent runs skip entirely.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

infra/k8s/argo/train-multi-seed-template.yaml

@@ -365,6 +365,7 @@ spec:
             if [ ! -x "$BINARY" ]; then
               echo "ERROR: precompute_features not found at $BINARY"; exit 1
             fi
+            export RAYON_NUM_THREADS=8
             if $BINARY \
               --data-dir /data/futures-baseline \
               --mbp10-data-dir /data/futures-baseline-mbp10 \