voice-qa-harness

Run an end-to-end smoke test against a Pipecat voice agent: spin up a candidate bot via voice-agent-qa, drive N turns, persist the transcript to the backend, fetch the resulting report, and diff against a known-good baseline. Surface regressions clearly. Use when the user says "smoke test the voice agent", "run nightly QA", "check the voice pipeline", "test the bot end-to-end", or similar.