Optimize mp4 parse times by not copying video data

[jprochazk]

• Part of Optimize mp4 parse times, and avoid duplicating their RAM use rerun#7481

Note that there's still a lot of optimization/refactor potential in this crate, which is why this doesn't close the above issue, but not copying all the video data is a great start.

Samples now have a helper which returns a range into the video data.

Breaking changes

• Removed Track::read_sample, superseded by Sample::range which can be used to index the original buffer
• Removed Track::data
• Sample byte offsets are no longer remapped in any way, they now represent an offset from the start of the MP4 file.
• read_file now returns a tuple of (Mp4, Vec<u8>) so that the original byte buffer is not lost and can be used together with Sample::range

[Wumpf]

nice!! Sounds like this gets us most of the way with minimal effort
That range method needs better doc though

ideally we'd have some form of shared ownership of the buffer, but don't know how that pans out with the arrow-owned buffer

[Wumpf]

thanks for going beyond! 🫶