Use cuda::std::numeric_limits in LAP kernels instead of passing infinity

[says1117]

Description

The LAP (Linear Assignment Problem) Hungarian-algorithm kernels took the sentinel
"infinity" value — std::numeric_limits<weight_t>::max() — as a host-computed
kernel argument. This was guarded by FIXME comments noting it should switch to
cuda::std::numeric_limits<weight_t>::max() once CUDA 10.2 was the baseline.

That baseline has long been met, so this PR resolves the FIXMEs by computing the
limit on-device instead of threading it through as a parameter.

Changes

• Drop the weight_t infinity parameter from kernel_rowReduction,
  kernel_columnReduction, kernel_dualUpdate_1, and kernel_dualUpdate_2.
• Use cuda::std::numeric_limits<weight_t>::max() directly inside those kernels
  (for the running-minimum initializers and the d_sp_min[spid] < infinity guard).
• Remove the corresponding host-side std::numeric_limits<weight_t>::max()
  arguments at the call sites in lap_functions.cuh.
• Add #include <cuda/std/limits> and delete the four resolved FIXME comments.

No functional or numerical behavior change — the sentinel value is identical; it is
now sourced on the device rather than passed from the host.

Testing

No new tests are required; this is a non-functional refactor covered by the existing
suite. Verified locally:

• SOLVERS_TEST builds clean.
• All 6 Raft.Hungarian* LAP tests pass (IntFloat, IntDouble, IntLong,
  LongFloat, LongDouble, LongLong).

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message 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: ee8b27b8-7e2c-4b29-aca5-6d339dc094f5

📥 Commits

Reviewing files that changed from the base of the PR and between aa9a6d5 and f28f810.

📒 Files selected for processing (2)

• cpp/include/raft/solver/detail/lap_functions.cuh
• cpp/include/raft/solver/detail/lap_kernels.cuh

---

📝 Walkthrough

Summary by CodeRabbit

• Refactor

  • Improved internal numerical handling in GPU-based assignment solving.
  • Simplified kernel interfaces while preserving existing solver behavior.

• Chores

  • Updated copyright information to include 2026.
  • Improved portability by using standard CUDA numeric-limit utilities.

Walkthrough

Changes

LAP sentinel cleanup

Layer / File(s)	Summary
Kernel sentinel logic cpp/include/raft/solver/detail/lap_kernels.cuh	Reduction and dual-update kernels remove explicit infinity parameters and use cuda::std::numeric_limits<weight_t>::max() internally.
Kernel launch updates cpp/include/raft/solver/detail/lap_functions.cuh	initialReduction and dualUpdate omit the removed infinity arguments when launching the updated kernels.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Suggested reviewers: vyasr, achirkin, cjnolet

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main change: moving LAP kernels to use cuda::std::numeric_limits instead of passing infinity.
Description check	✅ Passed	The description matches the changeset and objectives, covering the kernel parameter removal, device-side limits, and tests.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[says1117]

Could a maintainer please add the required labels when you get a chance? This is a non-functional refactor (identical sentinel value, just computed on-device), so improvement + non-breaking. Thanks!

[divyegala]

/ok to test 22ac277

[divyegala]

/merge