k8s(grid-rs): drop parallelism to 1 after 3 Gi still OOMs at 2

Two concurrent forecast tasks plus the NFS score-file write cache (23
files x ~2 MB) plus the analysis step's wgrib2 peak repeatedly crossed
the 3 Gi cgroup limit. Single-lane parallelism per pod keeps steady RSS
under 2 Gi; with 2 replicas that still gives 2 concurrent tasks
cluster-wide.

PROP_GRID_RS_PG_CONNS follows the parallelism+2 formula, dropping from
4 to 3.

k8s/deployment-grid-rs.yaml

@@ -82,17 +82,21 @@ spec:
               value: "/data/scores"
             - name: RUST_LOG
               value: "info"
-            # talos5 is 4c/4t (i5-7260U). 2 concurrent tasks per pod ×
-            # 2 replicas = 4 cluster-wide, enough to drain f01..f18 in
-            # ~5 min while keeping analysis (f00) room for its native-
-            # duct wgrib2 peak without OOM. Raising to 3 reintroduced
-            # the Stream-A-cutover OOM on the analysis step.
+            # Single concurrent task per pod. 2 replicas → 2 concurrent
+            # cluster-wide. Raised from 2 after the new 3 Gi limit still
+            # OOMed under 2-parallel forecast load — the NFS-backed
+            # score-file writes (23 files × ~2 MB each) keep cgroup page
+            # cache high enough that two concurrent tasks plus the
+            # wgrib2 working set regularly crosses 3 Gi. Single-lane
+            # parallelism stays comfortably under 2 Gi and lets the
+            # analysis step (native duct + NEXRAD + commercial merges)
+            # run safely within the same budget.
             - name: PROP_GRID_RS_PARALLELISM
-              value: "2"
+              value: "1"
             # Pool size = parallelism + 2 (NOTIFY + retry slack). Kept in
             # sync with PROP_GRID_RS_PARALLELISM when either changes.
             - name: PROP_GRID_RS_PG_CONNS
-              value: "4"
+              value: "3"
             - name: METRICS_ADDR
               value: "0.0.0.0:9100"
           envFrom: