[TensorRT EP] Enable a minimal CUDA EP compilation without kernels (#19052)
Adresses https://github.com/microsoft/onnxruntime/issues/18542.
I followed the advice given by @RyanUnderhill
[here](https://github.com/microsoft/onnxruntime/pull/18731#issuecomment-1848261925)
and went with a minimal CUDA EP for now.