Fix package loss for small models

[merrymercy]

Fix for #2548

fix packet loss when deploy small llm like qwen 0.5B or fishauido 

## Motivation

We found that when deploying some small LLMs, SGlang experiences packet loss under high load because the scheduler's results return too quickly, and at the same time, the rid is not updated in rid_to_state

## Modifications

we move send_to_scheduler after rid_to_state update

Co-authored with @sdli1995