refactor: standardize API response

refactor: process audio chunk for llm and frontend

fix: too many req

update websocket to get schema from context

rate limiter module for future scaling (not used)

add sliding window buffer, single writer pattern, hibrid trigger, dirty flag logic, and debouncing.

optimize websocket logic with debouncing and safety guards

update websocket to process audio by AIService every 250ms

update logic buffer to += and sliding windows for audioBuffer length > 60Mb

add websocket streaming handler