YUV image input

## ❓Question
Hi, I have videos in `yuv422p10le` format on which I have trained a PyTorch model. To reduce overhead I did not bother to convert from YUV to RGB I just convert the YUV values to `yuv444` and then to float and scale to interval 0.0-1.0, this works well.

Now I want to run this on live video on an iPhone with iOS18 and have a few questions about best practice and performance.
- Should I be using `coremltools.ImageType` for the input type? `color_layout` does not have an option for YUV
- I am guessing that video frames arrive in YUV format, is there any performance benefit of not casting to RGB?
- Should I just retrain my model on RGB images?

Thanks!

The issue names no repository file, test, or entry point; begin by locating ImageType/color_layout handling and the iOS input path. Done would require a maintainer-supported answer or a clearly scoped implementation request covering YUV input and its performance tradeoffs.

Written by the indexing model from the issue text.