Three fixes for audio delay issues caused by RTP timer drift

[gunarsritins]

Fix RTP timer drift causing audio delays

Three fixes for audio delay issues caused by RTP timer drift:

Changes

1. Periodic timer sync — every 250 ticks (~5s), resync the RTP timer to wall-clock before sleeping. Sub-tick scheduling latency compounds over time
   because timerfd read() returns 1 even when the timer fires late, causing gradual audio delay on long calls.

2. Hot socket burst prevention — when the RTP socket has queued packets, call timer_next() after timer_sync() so the timer sleeps for a full tick
   instead of returning immediately and bursting through queued packets. This prevents immediate audio delay spikes when packets queue up.

3. AUDIO_SYNC timer resync — when the media layer handles AUDIO_SYNC (codec changes, hold/resume), resync the RTP timer via new switch_rtp_timer_sync().
   Previously, resuming from hold could cause a burst of catch-up packets as the timer tried to make up for elapsed time, resulting in audio delay after
   hold/resume.

Testing

Verified on ARM64 (UDR and UCKP). Gradual drift on long calls, immediate delay spikes, and post-hold audio delay all resolved.

[AlexPereverzyevUbiquiti]

you can make it more efficient if the interval is power of 2, eg (rtp_session->timer.tick & 255) == 0