WebRTC + str0m server. Measures TTFT, TTFA, and turn-taking on the client by timing the
transcriptions data-channel events and the remote audio stream.
TTFT = user_final → first assistant_response text. TTFA = user_final → first audible remote audio. Gap = user_final → first audible remote audio (same as TTFA, kept for clarity in turn cards).
Speak after the connection goes live — turn cards appear here.