rtmp2: avoid stalling when stream starts with a high PTS
When a pipeline has been working for more than G_MAXINT32 but less than G_MAXUINT32 milliseconds (between 25 and 49 days), and we add rtmp2sink (e.g because remote rtmp server went up), code in buffer_to_message mistakes the high initial PTS for an underflow and forces PTS to 0 for this and all subsequent buffers, making RTMP stream unusable.