[FEA] CI should support HMM or Grace Hopper

[rongou]

Is your feature request related to a problem? Please describe.
We now have a system_memory_resource that allows system-allocated memory for device use. However, this only works on HMM or Grace Hopper, which the current CI pipelines don't support, so most of its tests are skipped.

Describe the solution you'd like
We need CI that supports HMM or GH. Probably the easiest way is to have a Turing GPU that runs open-source drivers (see https://developer.nvidia.com/blog/simplifying-gpu-application-development-with-heterogeneous-memory-management/).

Describe alternatives you've considered
If we could get a Grace Hopper machine for CI, that would be even better, but it's probably not very available in the near future.

Additional context
PR that added the memory resource: #1581

[jakirkham]

cc @ajschmidt8

[bdice]

This will be blocked until our new CI system is available. I'll mark it as blocked for now.

[ajschmidt8]

We'll have some new CI machines rolling out in the coming weeks.

The new machines will include L4s and H100s.

We have Grace Hopper machines in our inventory, but they require some additional investigation before they'll work correctly.