[coil-video] Support MediaDataSource implementations

[lukaspieper]

Is your feature request related to a problem? Please describe.
Currently Coil-Video only supports a small range of inputs (see MediaMetadataRetriever.setDataSource(...) and if I understand it correctly with a fallback that writes a whole file to disk that is way to slow for large files like videos. In my case the files are encrypted, therefor they cannot directly be used by coil.

Describe the solution you'd like
I have a MediaDataSource implementation that handles the decryption on the fly. However, I haven’t found a way to pass it down to coil without extending Coil's image pipeline by adding a class that is similar to VideoFrameDecoder but forwards the MediaDataSource to the MediaMetadataRetriever.

I just barely checked the code base, so there could be better solutions. The VideoFrameDecoder would need to recognize MediaDataSource to apply them directly. I think this could be done by adding an ImageSource that accepts MediaDataSource as parameter.

I have seen issue #25 so this new API would only be available on > API 23.

Additional context
While I was able to implement this myself by mostly copying VideoFrameDecoder, I think it would be nice to see support upstream.

[colinrtwhite]

Hmm this might be slightly tough to implement. An ImageSource implementation needs to be able to return a Source so it can be ready by any fetcher. A MediaDataSource could be wrapped as a Source, however when read by VideoFrameDecoder it would create a temporary file on disk, which we want to avoid. Additionally, we can't add this to the main artifact - it would need to exist as part of coil-video.

I think the best way to implement this is probably create a custom MediaDataSourceFetcher that wraps MediaDataSource as a Source-backed ImageSource and attaches a MediaSourceMetadata object, which can then be read by VideoFrameDecoder. Open to a PR that adds this!