Make UBSan warnings fatal and fix the existing ones

webrtc/modules/rtp_rtcp/source/rtp_sender.cc

Index: webrtc/modules/rtp_rtcp/source/rtp_sender.cc
diff --git a/webrtc/modules/rtp_rtcp/source/rtp_sender.cc b/webrtc/modules/rtp_rtcp/source/rtp_sender.cc
index 89e7735c7a4da9ae45dd42f0ed15ade497a28b79..2e227fdf4b7fd952d4ff16237a9e522fcea9e28f 100644
--- a/webrtc/modules/rtp_rtcp/source/rtp_sender.cc
+++ b/webrtc/modules/rtp_rtcp/source/rtp_sender.cc
@@ -925,7 +925,7 @@ void RTPSender::UpdateDelayStatistics(int64_t capture_time_ms, int64_t now_ms) {
     return;
 
   uint32_t ssrc;
-  int avg_delay_ms = 0;
+  double avg_delay_ms = 0;

[kwiberg-webrtc, 2017/09/03 02:39:47]

Again, int64_t?

[oprypin_webrtc, 2017/09/04 08:36:07]

Yeah makes sense. My logic was that `double` is safer for a much larger range of
values and still has exact precision for up to 52 bits. But int64 will work.

   int max_delay_ms = 0;
   {
     rtc::CritScope lock(&send_critsect_);
@@ -951,8 +951,8 @@ void RTPSender::UpdateDelayStatistics(int64_t capture_time_ms, int64_t now_ms) {
       return;
     avg_delay_ms = (avg_delay_ms + num_delays / 2) / num_delays;
   }
-  send_side_delay_observer_->SendSideDelayUpdated(avg_delay_ms, max_delay_ms,
-                                                  ssrc);
+  send_side_delay_observer_->SendSideDelayUpdated(
+      rtc::dchecked_cast<int>(avg_delay_ms), max_delay_ms, ssrc);
 }
 
 void RTPSender::UpdateOnSendPacket(int packet_id,