Add stream flag to io.read_video

The torchvision.io.read_video function does not have an option to exclude reading certain streams.

I would like to be able to load the frames of a video without loading audio. In order to do this I had to use internal functions like so:

container = av.open(fileath, metadata_errors='ignore')

info = {}
video_frames = []
video_frames = torchvision.io.video._read_from_stream(container, start, end,
                                      container.streams.video[0], {'video': 0})
info["video_fps"] = float(container.streams.video[0].average_rate)

Obviously this exposes the PyAV backend, therefore I envision two options:

• functions for video only, audio only, and both.
• optional parameters added to the read_video function.

Start at torchvision.io.read_video and the internal _read_from_stream path, then inspect how PyAV streams are selected. Done means the public API supports the requested stream-selection behavior, including video-only loading without audio, while retaining the existing combined behavior.