ffmpeg: honor -ss and -t parameters with muxed subtitles.

This patch fixes two things:

 - in case of subtitles, check_recording_time() is comparing the current
   PTS to the recording time (-t option, set to INT_MAX by default), so
   the -ss option needs to be taken into account. It is not required in
   do_{audio,video}_out() because this adjustment is set while polling
   the filtergraph (see poll_filters()).

 - It also adjusts the PTS sent to the encoder (and later transmitted to
   the muxer) so the TS in the output make sense and are not kept
   verbatim.

Note: this only works for muxers honoring the PTS, such as
lavf/matroskaenc. But for other such as the ASS muxer which just does a
verbatim copy, or the SubRip muxer which doesn't write the TS in some
cases, it will not work yet.

ffmpeg.c

@@ -1683,12 +1683,15 @@ static void do_subtitle_out(AVFormatContext *s,
     else
         nb = 1;
 
+    /* shift timestamp to honor -ss and make check_recording_time() work with -t */
+    pts = av_rescale_q(pts, ist->st->time_base, AV_TIME_BASE_Q)
+        - output_files[ost->file_index]->start_time;
     for (i = 0; i < nb; i++) {
-        ost->sync_opts = av_rescale_q(pts, ist->st->time_base, enc->time_base);
+        ost->sync_opts = av_rescale_q(pts, AV_TIME_BASE_Q, enc->time_base);
         if (!check_recording_time(ost))
             return;
 
-        sub->pts = av_rescale_q(pts, ist->st->time_base, AV_TIME_BASE_Q);
+        sub->pts = pts;
         // start_display_time is required to be 0
         sub->pts               += av_rescale_q(sub->start_display_time, (AVRational){ 1, 1000 }, AV_TIME_BASE_Q);
         sub->end_display_time  -= sub->start_display_time;