chore(infra): right-size api-private autoscaling

api-private was pinned at 4/4 max: it's a near-idle internal API (CPU
~1m) whose ~470Mi memory is a flat per-pod Node baseline, not
load-driven. Using memory as an HPA driver against an undersized 350Mi
request held utilization near the 150% target so the formula never
recommended scaling down. Drop memory from the HPA (CPU-only scaling;
memory safety is the 630 heap cap + 700 limit), raise minReplicas to 2
for HA, and right-size the request to 500Mi to match real usage. Settles
at ~2 pods instead of 4.

Author: idoshamun

.infra/index.ts

@@ -512,18 +512,18 @@ if (isAdhocEnv) {
         { name: 'ENABLE_PRIVATE_ROUTES', value: 'true' },
         ...jwtEnv,
       ],
-      minReplicas: 1,
+      minReplicas: 2,
       maxReplicas: 4,
       requests: {
-        memory: '350Mi',
+        memory: '500Mi',
         cpu: '10m',
       },
       limits: {
         memory: `${privateMemory}Mi`,
       },
       readinessProbe,
       livenessProbe,
-      metric: { type: 'memory_cpu', cpu: 200, memory: 150 },
+      metric: { type: 'memory_cpu', cpu: 200 },
       createService: true,
       service: {
         type: 'LoadBalancer',