(likely bug) WebRTC Video track constraints dimensions

In order to get a streamed input of NDArray[Shape["H, W, 3"]], the values for width and height needs to be flipped as follows:

```
input_img = WebRTC(label="Stream",
track_constraints={
"width": {"exact": H},
"height": {"exact": W},
},
rtp_params={"degradationPreference": "maintain-resolution"},
mode="send-receive",
modality="video",
mirror_webcam=False)
```

In the below example, I used H=960, W=720 (portrait orientation), following the flipped WebRTC track_constraints configuration as described above. It results in a fully formed portrait image (verified via cv2.imshow on the server), and a stream component/widget of the same portrait aspect ratio. However, the image rendered is consistently chopped off/offset up.


To summarize the issues:

1. flipped constraints
2. viewport cut-off

Start by reproducing the portrait case with H=960 and W=720, using the shown WebRTC track_constraints and checking both the server-side image and rendered stream widget. Trace how the Stream component applies track constraints and sizes its viewport; done means portrait input renders without requiring flipped width and height values and without cropping or offsetting the image.

Written by the indexing model from the issue text.