### What happened?

When the proxy reaches Redis through a proxy layer that blocks the `SCRIPT` command (Codis, Twemproxy, Kedis, and similar), multi-instance rate limiting silently stops working. Every pod falls back to its own in-memory counters, so a limit meant for the whole fleet is enforced per pod and the effective limit becomes the configured value times the number of pods. Nothing surfaces to the caller, so requests that should be throttled get through.

The rate limiter runs its sliding-window logic as a Lua script. On a fresh connection the script is not yet cached server side, so the first `EVALSHA` misses and redis-py reloads the script with `SCRIPT LOAD` before retrying. These Redis proxies forward `EVAL` and `EVALSHA` but reject `SCRIPT`, so the reload fails with `unknown command 'SCRIPT'`. The limiter treats that as a Redis outage and quietly drops to in-memory limits. The same registered-script path also backs distributed pod-lock release and the budget limiters, so those degrade the same way behind such a proxy.

Expected: behind a SCRIPT-blocking Redis proxy the Lua rate-limiter path keeps working and limits are enforced across all instances

Actual: the script c…