[GST][MSE] Make sure to extend only the very first sample when DTS = …

[asurdej-comcast]

…0 and PTS > 0.

In the absence of an edit list, ensure that WebKit’s AppendPipeline extends only a single sample.

This fixes the handling of invalid streams with multiple DTS = 0 samples. Modifying PTS values for all of them results in unexpected behavior, playback issues, PTS reordering, etc.

The problem was spotted in a video web application that uses TS -> MSE transmuxing.

[asurdej-comcast]

Alternatively, since Edit lists are supported, we can consider removing that code block or enable it for GST < 1.16 only

This was spot in AFV app with dailymotion/hls.js transmuxer. There were two samples with DTS=0 and PTS > 0

Jan 16 11:33:12 hisense-v2 HtmlApp-0[5557]: 0:00:15.350182055    26   0x931d20 DEBUG              webkitmse AppendPipeline.cpp:538:appsinkNewSample: Extending first sample of track '1' to make it start at PTS=0 buffer: 0xc77738, pts 0:00:00.023222222, dts 0:00:00.000000000, dur 0:00:00.000000000, size 37548, offset none, offset_end none, flags 0x4040
...
Jan 16 11:33:12 hisense-v2 HtmlApp-0[5557]: 0:00:15.350812617    26   0x931d20 DEBUG              webkitmse AppendPipeline.cpp:538:appsinkNewSample: Extending first sample of track '1' to make it start at PTS=0 buffer: 0x9f818800, pts 0:00:00.089888888, dts 0:00:00.000000000, dur 0:00:00.023222222, size 18350, offset none, offset_end none, flags 0x6000
...
Jan 16 11:33:12 hisense-v2 HtmlApp-0[5557]: [WPEWebKit:MediaSource:-] SourceBufferPrivateGStreamer::didReceiveSample(FEB30001) {"pts":{"value":0.056555,"numerator":56555,"denominator":1000000,"flags":1},"dts":{"value":0.023222,"numerator":23222,"denominator":1000000,"flags":1},"duration":{"value":0.033333,"numerator":33333,"denominator":1000000,"flags":1},"flags":0,"presentationSize":{"width":1920,"height":1080}}

[eocanha]

I think it's better to try to introduce this patch, since it may also benefit upstream, where edit list support is reverted (but should be eventually reintroduced).

The patch looks good, although having two (or more) samples with DTS=0 can cause other problems (regardless of this patch): Since both DTS are equal, we don't have much guarantees about the stability of the DecodeOrderSampleMap in SourceBufferPrivate holding the samples. Any of the two samples with DTS=0 could be returned earlier than the other one depending on storage ordering or just luck.

Patch submitted upstream for review as https://bugs.webkit.org/show_bug.cgi?id=305858 / WebKit/WebKit#56917.