Replace libcudf cuda stream owners

[vyasr]

Summary

• replace direct rmm::cuda_stream owners in libcudf tests and the FST benchmark with device-bound cuda::stream
• retain explicit synchronization while adopting non-blocking cuda::stream creation
• preserve the stream-pool examples and libcudf_streaming interfaces
• update stream-identification documentation for its cuda::stream owner

Validation

• built libcudf
• rebuilt libcudf_streaming
• passed focused AST, column rebind, scalar async, FST, logical-stack, streaming hash-join, and streaming groupby tests
• compiled the modified FST benchmark translation unit
• git diff --check
• verified remaining rmm::cuda_stream references are only deferred rmm::cuda_stream_pool examples

Notes

The local pre-commit suite passed all relevant checks; its Lychee hook is currently misconfigured because the configured revision is not a versioned release tag.

[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.