Fixes garbled videos in The Finals.
When cycling the video and restarting it from the beginning the game sets input and output types with the frame size not matching the video stream size. The transform doesn't (and should not) perform any scaling and the frame size is essentially ignored when comparing gstreamer caps, so wg_transform_set_output_format() doesn't do anything. That results in the disagreement between h264 decoder part in video_decoder.c and wg_transform on the frame size, so the smaller sample from wg_transform is copied to the larger one h264 decoder (wrongly) expects and results in garbled sample.
Windows (probably quite expectedly) changes stream format back in that case which results in MF_E_TRANSFORM_STREAM_CHANGE when getting next sample. The output sample frame size is only possible to change with SetOutputType if SetInputType is called before that (otherwise that just fails, we have a test for that). Windows also resets auto timestamp in this case (at least with input sample provided without timestamp), that is left todo in test.