Prepare scan-by-key tuning policy for lookahead

[bernhardmgruber]

Fixes part of: #10075

[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: def1c13f-f5ca-47f1-8eb2-ff125b133cb2

📥 Commits

Reviewing files that changed from the base of the PR and between 4d8580a and def9fa4.

📒 Files selected for processing (1)

• cub/cub/device/dispatch/tuning/tuning_scan_by_key.cuh

🚧 Files skipped from review as they are similar to previous changes (1)

• cub/cub/device/dispatch/tuning/tuning_scan_by_key.cuh

---

📝 Walkthrough

Summary by CodeRabbit

• Improvements
  • Updated scan-by-key tuning policy to introduce a dedicated lookback policy and an explicit algorithm selector.
  • Improved consistency between lookback settings and kernel launch/tile sizing, including related debug output.
  • Preserved scan-by-key behavior and existing tuning values.
• Tests
  • Updated scan-by-key device tests to validate the revised nested policy structure and updated policy string formatting.
• Chores
  • Reformatted a scan-by-key benchmark policy initializer.

Walkthrough

Changes

The scan-by-key tuning policy now contains an explicit algorithm discriminator and nested lookback configuration. Dispatch, benchmarks, and tests use the nested fields for policy selection, kernel setup, tile sizing, and serialization.

Scan-by-key policy refactor

Layer / File(s)	Summary
Policy model and selection cub/cub/device/dispatch/tuning/tuning_scan_by_key.cuh	Adds ScanByKeyLookbackPolicy, nests it in ScanByKeyPolicy, and updates policy conversion and compute-capability selection.
Dispatch and kernel configuration cub/cub/device/dispatch/dispatch_scan_by_key.cuh	Reads launch bounds, agent parameters, tile sizes, logging, and launcher arguments from policy.lookback.
Policy consumers and validation cub/benchmarks/bench/scan/exclusive/by_key.cu, cub/test/catch2_test_device_scan_by_key_env.cu, cub/test/catch2_test_device_scan_by_key_env_api.cu	Updates policy initializers, block-size access, designated members, and serialized expectations.

Possibly related PRs

• NVIDIA/cccl#10077: Refactors a related tuning policy into an algorithm discriminator with nested lookback parameters.
• NVIDIA/cccl#10081: Applies the same nested lookback policy pattern to dispatch, launch configuration, and tests.

Suggested reviewers: shwina, pauleonix

---

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

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 1h 57m: Pass: 100%/287 | Total: 3d 22h | Max: 1h 24m | Hits: 96%/209463

See results here.