[llm.serving] Reconfigure router to better perform under high concurrency

[GeneDer]

Why are these changes needed?

Accordingly to internal benchmarks the router deployment becomes bottleneck under high concurrency situation. This PR reconfigured the router to use a shorter target ongoing request (16) to better scales the router. And to make router's min_replicas, initial_replicas, and max_replicas to be a factor of N out of all the models. The multiplication factor can also be configured via env var RAYLLM_ROUTER_TO_MODEL_REPLICA_RATIO and is defaulted to 2 (meaning there will be roughly 2 router replicas for every model replica)

Benchmark plot:
Fixed number of router replicas + different amount of model replicas

Variable number of router replicas + model replicas

Related issue number

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
  • I've added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(