fix: work around RTCP SR not sent for all SSRCs

[getroot]

Per the RTP/RTCP spec, a sender must send RTCP SR for every SSRC it uses. However, some RTP-based sources only send SR for a subset of their SSRCs.

This issue was observed with OBS 32.1.0 sending WHIP Simulcast.

This patch works around the issue by not starting until RTCP SR is received for all registered SSRCs, and reduces the fallback timeout from 5 seconds to 3 seconds so it falls back to per-track timestamp calculation without RTCP SR more quickly.

---

Update: Also made LipSyncClock thread-safe in preparation for potential future multi-threaded use. Currently RTP and RTCP are always handled on the same thread, so there is no practical issue today.

[getroot]

I have improved the LipSyncClock library so that it works well in a multi-threaded environment. (Although this is not an issue since it is currently used in a single-threaded environment,) please review it again.
@dimiden

[dimiden]

I have improved the LipSyncClock library so that it works well in a multi-threaded environment. (Although this is not an issue since it is currently used in a single-threaded environment,) please review it again. @dimiden

If clock->_updated = true is set in UpdateSenderReportTime() and then CalcPTS() evaluates if (clock->_updated == false) right after, could this lead to unexpected results when calculating pts?
(If this level of discrepancy does not have a meaningful impact, please feel free to disregard this comment.)

[getroot]

@dimiden That is a good point. Considering multithreading, the clock must be atomic. I have patched this.

[dimiden]

LGTM 👍