fix(webapp): default LLM_PRICING_RELOAD_PUBSUB_ENABLED to off

Most processes that import the registry (dashboard, workers, single-
service self-hosted webapp) don't actually need real-time pricing
freshness — the existing 5-minute interval is fine. Only OTel-ingesting
services where pricing directly affects span cost enrichment need to
subscribe.

Default off, opt-in on the span-ingesting services in multi-service
deployments. Self-hosters running a single webapp can flip it on if
they want sub-second freshness, but the default keeps reload load
off processes that don't benefit from it.

Author: ericallam

apps/webapp/app/env.server.ts

@@ -1427,11 +1427,11 @@ const EnvironmentSchema = z
     LLM_PRICING_RELOAD_CHANNEL: z.string().default("llm-registry:reload"),
     LLM_PRICING_RELOAD_DEBOUNCE_MS: z.coerce.number().int().default(1000),
     // Whether to subscribe this process to the LLM_PRICING_RELOAD_CHANNEL.
-    // Defaults to true so single-service self-hosted deployments work without
-    // tuning. In multi-service deployments, set this to false on services
-    // that don't ingest spans (dashboard, workers) — only the OTel-ingesting
-    // services need the registry to reload in real time.
-    LLM_PRICING_RELOAD_PUBSUB_ENABLED: BoolEnv.default(true),
+    // Default off — only OTel-ingesting services need real-time pricing
+    // freshness; dashboard/worker processes are fine on the existing
+    // 5-minute periodic reload. In multi-service deployments, set this to
+    // true on the span-ingesting services.
+    LLM_PRICING_RELOAD_PUBSUB_ENABLED: BoolEnv.default(false),
     LLM_PRICING_SEED_ON_STARTUP: BoolEnv.default(false),
     LLM_PRICING_READY_TIMEOUT_MS: z.coerce.number().int().default(500),
     LLM_METRICS_BATCH_SIZE: z.coerce.number().int().default(5000),

apps/webapp/app/v3/llmPricingRegistry.server.ts

@@ -37,11 +37,11 @@ export const llmPricingRegistry = singleton("llmPricingRegistry", () => {
     });
   }, reloadInterval);
 
-  // Pub/sub reload is opt-in per process. Without it, the registry stays
-  // accurate via the existing 5-minute interval. In multi-service deployments
-  // we only want the OTel-ingesting services subscribed — the dashboard and
-  // worker services don't need real-time pricing freshness and shouldn't pile
-  // onto each publish with a full-table reload.
+  // Pub/sub reload is opt-in per process (default off). Without it, the
+  // registry stays accurate via the existing 5-minute interval. Enable on
+  // the OTel-ingesting services where pricing freshness directly affects
+  // span cost enrichment; dashboard and worker services don't need it and
+  // shouldn't pile onto each publish with a full-table reload.
   if (env.LLM_PRICING_RELOAD_PUBSUB_ENABLED) {
     const subscriber = createRedisClient("llm-pricing:subscriber", {
       keyPrefix: "llm-pricing:subscriber:",