feat(cuda): add canonical GELU and sigmoid providers

[voltjia]

Summary

• Add canonical Gelu and Sigmoid CUDA providers for NVIDIA, Iluvatar, MetaX, and Moore.
• Reuse the existing CUDA-compatible kernels and add focused Gelu coverage for both supported approximation modes.

Motivation

InfiniCore still uses the deprecated GeluInfinilm, GeluTanhInfinilm, and SigmoidInfinilm compatibility operators. Canonical provider coverage is required before those adapters can migrate to the PyTorch-aligned operator names.

Type of Change

• feat - new feature / new operator / new platform
• fix - bug fix
• perf - performance improvement (no behavioral change)
• 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 (requires a ! in the Conventional Commits prefix or a BREAKING CHANGE: footer)

Platforms Affected

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

Smoke Test Result

Validated on ssh nvidia in accelerator-dev/nvidia:latest as part of the canonical-provider integration stack:

$ scripts/dev/build.sh nvidia --jobs 16
Build succeeded

$ python -m pytest tests/test_gelu.py tests/test_sigmoid.py --devices nvidia -q
148 passed

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	build passed	148 focused tests passed
Iluvatar	Yes	not run	platform hardware unavailable; platform CI required
MetaX	Yes	not run	platform hardware unavailable; platform CI required
Cambricon	No	N/A	N/A - not affected
Moore	Yes	not run	platform hardware unavailable; platform CI required
Ascend	No	N/A	N/A - not affected

Benchmark / Performance Impact

N/A. The canonical providers reuse existing kernels.

Notes for Reviewers

The public schemas already existed; this PR adds native provider coverage and the metadata needed by the generic implementations.

InfiniOps API	Alignment target	Evidence
Gelu(input, approximate, out)	PyTorch torch.nn.functional.gelu(input, approximate='none')	PyTorch GELU documentation
Sigmoid(input, out)	PyTorch torch.sigmoid(input, *, out=None)	PyTorch sigmoid documentation

InfiniOps uses an explicit trailing output tensor for its C++ API. This PR does not add overloads or remove the deprecated InfiniLM compatibility classes. The vendor backend headers are thin wrappers around the shared CUDA implementations.

The NVIDIA result validates the NVIDIA provider and shared kernels. Iluvatar, MetaX, and Moore platform CI remain required.