RSDK-14053 Add Conversation timeout attribute

[oliviamiller]

Summary

• Adds a conversation_timeout_seconds config attribute. After a wake word fires and a segment is yielded, any speech that begins within this window streams through without requiring the wake word again — letting users ask follow-up questions naturally.
• Sliding window: each yielded segment refreshes the timer, so the conversation stays open as long as the user keeps talking. After the timer expires, the wake word is required again.
• Works for both OWW (engages on IDLE→ACTIVE transition, streams live) and Vosk (bypasses transcript matching in _run_detection).
• Default is 0 (disabled) — module behaves as a strict wake-word filter when not configured.

Implementation

• New _in_conversation_window() / _refresh_conversation_window() helpers using time.monotonic().
• OWW path: hooks into _process_vad_frame at IDLE→ACTIVE to set oww_detected = True directly, so the existing streaming path engages without an OWW inference hit.
• Vosk path: branches inside _run_detection to yield buffered chunks + sentinel unconditionally during the window.
• Refresh fires after every sentinel emit (streaming path, OWW batch path, Vosk match).

Testing

• 15 new unit tests: validation, init, helper behavior, OWW in/out of window, Vosk bypass, sliding refresh.
• All 112 tests pass, lint clean.

[seanavery]

Is this new python syntax? Do we need to update anything in pyinstaller or CI to make sure we are building with newer python ver?

[oliviamiller]

needs python 3.10+, ill update setup.sh to make that explicit

[seanavery]

Do we need min_speech_frames check on the oww path?

[oliviamiller]

min_speech_frames in only meaningful for vosk, meant to avoid running inference on speech segments that are too small

[seanavery]

LGTM