[alexluong]

rsmq loads its Lua scripts once in NewRedisSMQ and then calls them only by hash with EVALSHA. A Redis or Dragonfly restart empties the server's script cache, so every receive, pop and visibility change fails with NOSCRIPT until Outpost restarts. For the retry scheduler: ~1 min of error backoff, then max consecutive errors reached, the retrymq-consumer worker stops, /healthz returns 503 and no retries run until the process restarts. Queued retries aren't lost; they wait.

This PR runs the scripts through go-redis Script.Run: EVALSHA, falling back to EVAL (which reloads the cache) on NOSCRIPT.

1. RedisClient: EvalSha + ScriptLoad replaced by RunScript(*redis.Script, keys, args...), used by receive, pop and change visibility.
2. No SCRIPT LOAD at startup; the first call loads each script.
3. TestScriptCacheFlushed: SCRIPT FLUSH before each script call. Fails with NOSCRIPT on main.

🤖 Generated with Claude Code