Required channels not respected on predict

[gnodar01]

Related to piximi/piximi#536.

The Model class lets subclasses define a required number of channels: https://github.com/piximi/piximi/blob/master/src/utils/common/models/Model.ts#L15

After adding the stardist-flourscence model in piximi-beta piximi/piximi#3, the requiredChannels was set to 1. When pressing predict on an image with 3 channels, piximi does not try to stop the user. Instead it attempts to pass the image in for prediction, and fails with a shape mismatch error from tensorflow.js.

What piximi should do in case of channel mismatch will probably vary.

• It should at least issue a warning, and not try to predict on an image with a channel mismatch from the model.
• A more sophisticated approach would be, If the number of channels the image has is greater than the number of channels required by the model, it will ask which channel(s) to use.
• If the image is one channel, and the model takes n channels, we may automatically (or with permission) duplicate the 1 channel, n times.