[Bug]: Console stats logging is incorrect when using api-server scaleout

[njhill]

StatsLoggers run in the front-end process. When there are multiple of these they will each be processing a subset of requests and thus have incomplete stats.

The prometheus stats logger aggregates the stats via it's shm-based multiprocessing mode, but the periodic console log that shows request queue lengths, throughput, and kv cache usage will happen independently in each front-end process and thus be incomplete/incorrect.

At a minimum we should disable this statslogger for api_server_count > 1.

[chaunceyjiang]

I can try it.

[njhill]

Thanks @chaunceyjiang!