Restructure the RLE encode tuning policy

[nanan-nvidia]

This is taken out from #9699 so that we can ship the changes to the tuning APIs first.

The changes are:

// before
_CCCL_HOST_DEVICE_API constexpr auto operator()(cuda::compute_capability) const -> cub::RleEncodePolicy
{
  return {256, 14, cub::BLOCK_LOAD_DIRECT, cub::LOAD_CA, cub::BLOCK_SCAN_WARP_SCANS, {}};
}

// after
_CCCL_HOST_DEVICE_API constexpr auto operator()(cuda::compute_capability) const -> cub::RleEncodePolicy
{
  return {cub::RleAlgorithm::lookback,
          {256, 14, cub::BLOCK_LOAD_DIRECT, cub::LOAD_CA, cub::BLOCK_SCAN_WARP_SCANS, {}}};
}

After this is merged, #9699 needs to be rebased from here.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 51183fec-c718-4d48-90ac-3d261be7a96e

📥 Commits

Reviewing files that changed from the base of the PR and between c3a18e7 and cb660f6.

📒 Files selected for processing (4)

• cub/benchmarks/bench/run_length_encode/encode.cu
• cub/cub/device/device_run_length_encode.cuh
• cub/cub/device/dispatch/tuning/tuning_rle_encode.cuh
• cub/test/catch2_test_device_run_length_encode_env.cu

---

📝 Walkthrough

Summary by CodeRabbit

• New Features

  • Added explicit algorithm selection for run-length encoding policies, currently supporting the lookback algorithm.
  • Added structured lookback tuning settings for clearer policy configuration and reporting.

• Improvements

  • Updated device run-length encoding to use the new policy structure.
  • Preserved compute-capability-specific tuning behavior while improving policy organization and diagnostics.

Walkthrough

Changes

The RLE tuning policy now carries an explicit lookback algorithm and nested lookback configuration. Compute-capability selection, device adaptation, benchmarks, and tests use the updated structure.

RLE policy refactor

Layer / File(s)	Summary
Policy types and serialization cub/cub/device/dispatch/tuning/tuning_rle_encode.cuh	Adds RleAlgorithm and RleLookbackPolicy, nests lookback tuning in RleEncodePolicy, and updates comparisons and streaming.
Compute-capability tuning selection cub/cub/device/dispatch/tuning/tuning_rle_encode.cuh	Returns lookback sub-policies for SM-specific cases and wraps them with RleAlgorithm::lookback.
Adapter and policy consumers cub/cub/device/device_run_length_encode.cuh, cub/benchmarks/bench/run_length_encode/encode.cu, cub/test/catch2_test_device_run_length_encode_env.cu	Updates policy adaptation, benchmark construction, test fixtures, aggregate initialization, and expected serialized output.

Suggested reviewers: bernhardmgruber, gonidelis

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 1h 59m: Pass: 100%/287 | Total: 2d 09h | Max: 1h 54m | Hits: 78%/257856

See results here.