Add a generic decode_image in tensorflow/io

[yongtang]

This is more or less a request from https://github.com/tensorflow/tensorflow/issues/18250#issuecomment-622265251

In tensorflow/io, a wider range of image formats have been added. Though it is still not possible to wire up those additional supported formats in a generic decode_image way like TF. The progress will be tracked here.

One thing to take into consideration, is the output of decode_image. While jpeg/png/webp are single frame formats, there are GIF/TIFF/OpenEXR that could include frames with even different shape. This will need to be taken into consideration. (TF's decode_image is an example where GIF is not handled very well due to the output shape).

[gtoderici]

FYI - WebP also supports animation, so that should also be taken into account.