Use cudaMemcpyBatchAsync

[bdice]

Description

This updates libcudf to use cudaMemcpyBatchAsync on supported systems (CUDA 12.8+). This API can be lower-overhead and may reduce multi-thread contention compared to cudaMemcpyAsync.

TODO:

• [ ] Fix errors -- some tests need synchronizations where we were missing them before
• [ ] Document that we prefer to use cudf utilities for memcpy inside libcudf

Checklist

• [x] 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[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.