Account for video with DTS shift and resulting negative dts values

[Wumpf]

This fixes

• Problem time-stepping some videos rerun#8098

Other mp4 parsers/data processors apply similar things:

• ffmpeg does, see code comments
• this one does https://github.com/kaltura/nginx-vod-module/blob/26f06877b0f2a2336e59cda93a3de18d7b23a3e2/vod/mp4/mp4_parser.c#L1205
• mp4 box does not!

... still this begs the question:

• is this speced? couldn't find anything
• this blurs the line a bit of what "the raw PTS/DTS" is. Rerun already offsets by min PTS. Maybe we should do that as well inside re_mp4. This way, ffprobe's output would always match ours (it does not right now as it's shifted by the min PTS, see example below)

First 6 frames in Rerun of the video from rerun-io/rerun#8098:

And for comparison the same frames from ffprobe, using ffprobe -of json -show_packets -select_streams v ../internal-test-assets/dtsbroken.mp4 > dtsbroken.json:

 {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 0,
            "pts_time": "0.000000",
            "dts": -2002,
            "dts_time": "-0.066733",
            "size": "51711",
            "pos": "24912",
            "flags": "K__"
        },
        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 4004,
            "pts_time": "0.133467",
            "dts": -1001,
            "dts_time": "-0.033367",
            "size": "7564",
            "pos": "79488",
            "flags": "___"
        },
        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 2002,
            "pts_time": "0.066733",
            "dts": 0,
            "dts_time": "0.000000",
            "size": "2192",
            "pos": "87052",
            "flags": "___"
        },
        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 1001,
            "pts_time": "0.033367",
            "dts": 1001,
            "dts_time": "0.033367",
            "size": "1911",
            "pos": "89244",
            "flags": "___"
        },
        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 3003,
            "pts_time": "0.100100",
            "dts": 2002,
            "dts_time": "0.066733",
            "size": "1168",
            "pos": "91155",
            "flags": "___"
        },
        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 8008,
            "pts_time": "0.266933",
            "dts": 3003,
            "dts_time": "0.100100",
            "size": "7874",
            "pos": "94674",
            "flags": "___"
        },

Note that this matches now for this video!

---

However, on "timestamped bunny" we still have the min pts discrepancy which we account for upon query in Rerun (unchanged from before/after this PR!):

        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 0,
            "pts_time": "0.000000",
            "dts": -512,
            "dts_time": "-0.033333",
            "duration": 256,
            "duration_time": "0.016667",
            "size": "190349",
            "pos": "48",
            "flags": "K__"
        },
        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 1024,
            "pts_time": "0.066667",
            "dts": -256,
            "dts_time": "-0.016667",
            "duration": 256,
            "duration_time": "0.016667",
            "size": "13884",
            "pos": "190397",
            "flags": "___"
        },
        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 512,
            "pts_time": "0.033333",
            "dts": 0,
            "dts_time": "0.000000",
            "duration": 256,
            "duration_time": "0.016667",
            "size": "3790",
            "pos": "204281",
            "flags": "___"
        },
        {
            "codec_type": "video",
            "stream_index": 0,
            "pts": 256,
            "pts_time": "0.016667",
            "dts": 256,
            "dts_time": "0.016667",
            "duration": 256,
            "duration_time": "0.016667",
            "size": "2169",
            "pos": "208071",
            "flags": "___"
        },

It works, so I'd say we leave it for now 🤔 . But something that might came back to haunt us!

[emilk]

this blurs the line a bit of what "the raw PTS/DTS" is. Rerun already offsets by min PTS. Maybe we should do that as well inside re_mp4. This way, ffprobe's output would always match ours (it does not right now as it's shifted by the min PTS, see example below)

Yeah, it would be nice if re_mp4 did this so that complexity was moved out of Rerun 🤔