fix: semaphore cegah backlog inference, timeout 5s

fix: Groq explicit timeout, prevent 40s hang

optimize: buffer 1.2s, overlap 50%, timing log Groq

feat: tambah timing log Groq

optimize: buffer 1.5s + silence detection

optimize: buffer 1.5s untuk kurangi latency

fix: is_active stop thread ONNX saat pipeline stop

feat: parallel ONNX+Groq inference

optimize: buffer 2s, groq timeout 3s, overlap 33%

fix: threshold -0.5 + clear buffer saat pause

fix: hapus file_name parameter

fix: load processor dari openai/whisper-small

fix: load processor dari HF Hub

fix: rename asr dan requirements