Add cuda_async_managed_memory_resource.

[bdice]

Description

Draft! Not yet ready for feedback. Contributes to #2054.

TODO:

• Look over C++ tests and make sure they're providing sufficient coverage (AI generated)
• Python bindings
• Docs
• Decide whether to reimplement managed_memory_resource on CUDA 13 with this?
• Determine whether decompression engine flags work, if provided (if so, should we use the default pool?)
• Determine whether we need to implement a release threshold argument, or provide docs on how to set that.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[bdice]

There is parallel design work happening in NVIDIA/cccl#5998. I don't want to offer this as a stable API in RMM given that we are fairly close to being able to use CCCL's MR implementations directly. I will mark this as experimental.