fix: fail Redis commands fast instead of hanging the request

With REQUIRE_API_KEY turned off, /price stopped answering at all — no
status, no error, just a timeout. /status kept working, which made it
look like the key change had half-applied.

ioredis queues commands while disconnected by default, so against a host
that no longer resolves every call sits in the offline queue waiting for
a connection that never arrives. The await never settles. That is a hang
rather than a rejection, so the try/catch around getCachedPrice cannot
see it and the fallback to Postgres it exists to trigger never runs.

Nothing new broke here. The hang has been present for as long as Redis
has been gone; the 401 was hiding it, because the auth hook rejected
every request before a handler could reach Redis. Opening the API up
removed the thing that was accidentally protecting it.

enableOfflineQueue: false makes those commands reject immediately, which
is what every caller already assumes: cached reads fall back to the
database, cache writes are best-effort, and x402 metering fails closed
on a request it cannot meter.

Worth noting the shape of this one — a dead dependency that produced no
errors, no logs and no failed requests, only silence, and became visible
solely because an unrelated setting stopped masking it.

Claude-Session: https://claude.ai/code/session_01USgemLt4Rnz4SGB1Srf3GB

Author: Miracle656

src/redis.ts

@@ -4,6 +4,21 @@ import { config, activeNetwork } from './config'
 export const redis = new Redis(config.redis.url, {
   maxRetriesPerRequest: 3,
   lazyConnect: true,
+  // Fail commands immediately while disconnected instead of queueing them.
+  //
+  // ioredis defaults this to true, so with an unreachable host every command
+  // sits in the offline queue waiting for a connection that never comes. The
+  // await simply never resolves — a hang, not an error, so the try/catch in
+  // getCachedPrice cannot help and the request dies of timeout instead of
+  // falling through to Postgres.
+  //
+  // It stayed hidden while REQUIRE_API_KEY was on, because the auth hook
+  // rejected requests before any handler could reach Redis. Opening the API up
+  // turned a 401 into a hang, which looked like the key change had failed.
+  //
+  // Rejecting fast is what the callers already expect: reads fall back to the
+  // database, writes are best-effort, and x402 metering fails closed.
+  enableOfflineQueue: false,
   // ioredis retries about once a second forever by default. Against a host
   // that no longer resolves that is a DNS lookup and two log lines every
   // second — roughly 170k lines a day, which buries every real message in the