Migrate native streams to cuda::stream_ref

[vyasr]

Summary

• Migrate JNI native stream parameters from rmm::cuda_stream_view to cuda::stream_ref.
• Replace stream handle/synchronization calls with stream.get() and stream.sync() for the new API.
• Replaces the conflicted branch in Migrate from rmm::cuda_stream_view to cuda::stream_ref #5014 on top of latest main.

Closes NVIDIA/cudf-spark#15685

Related to NVIDIA/cudf#23636 and NVIDIA/cudf#23770.

[greptile-apps]

Greptile Summary

The PR mechanically migrates native CUDA stream interfaces from rmm::cuda_stream_view to cuda::stream_ref while preserving existing GPU operations.

• Updates native declarations, definitions, helpers, and kernel launches to use cuda::stream_ref.
• Replaces raw stream access and synchronization with get() and sync().
• Updates native tests and NVBench stream adapters to compile against the migrated interfaces.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
src/main/cpp/src/bloom_filter.cu	Migrates Bloom filter stream parameters, synchronization, and kernel-launch handle access to the new CUDA stream API.
src/main/cpp/src/hash/hash.hpp	Updates the shared hashing API declarations, including SHA wrappers, to accept cuda::stream_ref.
src/main/cpp/src/row_conversion.cu	Migrates row-conversion allocation, copy, and kernel operations to the new stream reference type.
src/main/cpp/src/protobuf/protobuf_kernels.cu	Updates Protobuf kernel and helper stream plumbing without changing parsing or output behavior.
src/main/cpp/src/shuffle_split.cu	Migrates shuffle-splitting stream propagation and raw CUDA handle access while retaining existing execution flow.
src/main/cpp/benchmarks/bloom_filter.cu	Adapts the benchmark to obtain the raw stream with get() and synchronize through sync().

_{Reviews (3): Last reviewed commit: "Merge main into stream_ref branch" | Re-trigger Greptile}

[vyasr]

This is a current-main replacement for the now-conflicting #5014. It carries the native stream migration needed after NVIDIA/cudf#23770 changes cuDF stream APIs to cuda::stream_ref.

All checks on this PR are green. Once this merges, we can rerun the remaining spark-rapids-jni job on NVIDIA/cudf#23770.

[vyasr]

My agent got trigger-happy and opened this. I will close for now in favor of #5014, but we can reopen this or cherry-pick changes from here onto #5014 as desired.