Introduce mechanism to extract image/video metadata without actually decoding frame data

[rmk]

Currently ffmpeg and all other tools relying on avformat_find_stream_info to extract metadata for some codecs always decode an entire frame in cases where theoretically this would not be necessary, examples being JPEG, PNG and TIFF but there are probably more. This makes avformat_find_stream_info very inefficient for that purpose in these cases (e.g. having to decode a 20 megapixel jpg or a 400MB tiff image to find out the image dimensions is orders of magnitude slower than it could be if only headers were parsed).

This was discussed on the mailing list in this thread including suggestions for a solution approach:

​http://ffmpeg-users.933282.n4.nabble.com/Probing-jpg-png-and-tiff-images-without-decoding-frames-td4660508.html