[REQUEST] Support running GPU kernels with arrays in system memory on Grace Hopper

[shi-eric]

Description

It should be possible to mix-and-match arrays on Grace Hopper systems, e.g. CPU kernels can access arrays residing in GPU memory and vice versa. A runtime check requiring all arrays to be allocated on the same device as the kernel is one of the things preventing this from working:

warp/warp/context.py

Lines 4945 to 4949 in 211b962

	# check device
	if value.device != device:
	raise RuntimeError(
	f"Error launching kernel '{kernel.key}', trying to launch on device='{device}', but input array for argument '{arg_name}' is on device={value.device}."
	)

There might be additional changes required for this capability to be performant.

Similarly, it should be possible to mix-and-match GPU arrays when peer access is enabled on peer-capable GPUs.

Context

Take advantage of platform features.