Make ImageSegmentation preprocess consistent with other CV models

[pweglik]

In ImageSegmentation.cpp we do:

 cv::Size inputSize = input.size();

  cv::resize(input, input, modelImageSize);

  std::vector<float> inputVector = imageprocessing::colorMatToVector(input);
  return {
      executorch::extension::make_tensor_ptr(getInputShape()[0], inputVector),
      inputSize};

but in other models we do:

return {imageprocessing::getTensorFromMatrix(getInputShape()[0], image),
          originalSize};

From my understanding those are equivalent and using util function is great, so let's keep our implementations consistent. Btw it looks like it could be one util function and not a method - we duplicate a lot of code this way, which causes problems in cases like this