feat(linked): add vLLM grouped_topk provider

[voltjia]

Summary

• Add a linked NVIDIA slot 16 for vLLM _moe_C::grouped_topk.
• Preserve the existing native slot 0 and public InfiniOps API.
• Cover native slot 0 and vLLM slot 16 in the existing grouped_topk test module.

Motivation

InfiniOps currently carries a native grouped_topk implementation derived from
upstream serving kernels. The installed vLLM provider already exposes the
operation through the C10 Dispatcher, so this PR adds a thin linked adapter
without copying another provider kernel into the repository.

#906 is merged and provides the shared vllm/_moe_C.*.so library mapping.

N/A - no linked issue.

Type of Change

• feat - new feature / new operator / new platform
• fix - bug fix
• perf - performance improvement
• refactor - code restructuring without behavior change
• test - adding or fixing tests only
• docs - documentation only
• build / ci - build system or CI configuration
• chore - tooling, formatting, or other non-code changes
• Breaking change

Platforms Affected

• CPU (WITH_CPU)
• NVIDIA (WITH_NVIDIA)
• Iluvatar (WITH_ILUVATAR)
• MetaX (WITH_METAX)
• Cambricon (WITH_CAMBRICON)
• Moore (WITH_MOORE)
• Ascend (WITH_ASCEND)
• Hygon (WITH_HYGON)
• PyTorch C++ bindings (WITH_TORCH)
• Build system / CMake / CI
• Python bindings / user-facing API

Smoke Test Result

Remote environment: A100 with the Cornserve vLLM provider image, using an
operator-pruned Release build with WITH_LINKED=ON.

python -m pytest tests -m smoke --devices nvidia -q
79 passed, 19 skipped, 7807 deselected

The smoke allowlist is unchanged.

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	Release operator-pruned build and existing smoke passed on A100	Linked slot 16: 4 passed; native slot 0: 13 passed, 1 skipped
Iluvatar	No	N/A	N/A
MetaX	No	N/A	N/A
Cambricon	No	N/A	N/A
Moore	No	N/A	N/A
Ascend	No	N/A	N/A

Benchmark / Performance Impact

N/A - no performance claim.

Notes for Reviewers

• The adapter computes the vLLM provider inputs required by the registered
  schema (scores and scores_with_bias), calls the provider on the caller's
  stream, and copies the returned tuple into InfiniOps' output-tail tensors.
• Slot 16 intentionally follows the vLLM provider's observable low-precision
  rounding, tie, and nonfinite behavior. It does not claim bitwise equivalence
  with native slot 0. The native implementation remains the default and keeps
  the existing InfiniOps contract.
• Provider-specific cases in tests/test_grouped_topk.py explicitly select slot 16
  and compare against the loaded vLLM operator. Without that provider, only those
  cases skip; the native regression cases still run.