Avoid installing onnxruntime_directml on windows

[apsonawane]

Currently when running tkml on windows it by default installs onnxruntime_directml, but if the scenario is for cuda on windows then we do not need onnxruntime_directml.

In CUDA scenario onnxruntime_gpu need to be installed.

Please look for a workaround in this case as installing both the packages may cause issues while running the command