refactor(cuda): adapt InfiniLM GELU and sigmoid providers

[voltjia]

Summary

• Replace the standalone CUDA implementations behind GeluInfinilm, GelutanhInfinilm, and SigmoidInfinilm with thin adapters to the canonical providers introduced by feat(cuda): add canonical GELU and sigmoid providers #889.
• Preserve the deprecated InfiniLM schemas and their existing tests; this PR changes provider ownership only.

Motivation

The deprecated compatibility operators should not retain separate CUDA kernels after canonical, open-source-aligned operators exist. Sharing the providers keeps legacy callers working while removing duplicate implementation paths.

Depends on #889.

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

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

cmake -S . -B build-adapters -DINFINI_OPS_OPS=gelu,gelu_infinilm,gelutanh_infinilm,sigmoid,sigmoid_infinilm
cmake --build build-adapters --parallel 8
# NVIDIA build passed

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	focused build passed	247 passed, 173 skipped
Iluvatar	No	N/A - not affected	N/A
MetaX	No	N/A - not affected	N/A
Cambricon	No	N/A - not affected	N/A
Moore	No	N/A - not affected	N/A
Ascend	No	N/A - not affected	N/A

Focused pytest output

python3 -m pytest tests/test_gelu.py tests/test_gelu_infinilm.py tests/test_gelutanh_infinilm.py tests/test_sigmoid.py tests/test_sigmoid_infinilm.py -q
247 passed, 173 skipped in 1.71s

Benchmark / Performance Impact

N/A. This is provider reuse with no intended behavior or performance change.

Notes for Reviewers

Deprecated compatibility API	Canonical target	Alignment basis
GeluInfinilm(input, approximate, out)	Gelu(input, approximate, out)	PyTorch torch.nn.functional.gelu(input, approximate='none')
GelutanhInfinilm(input, out)	Gelu(input, tanh, out)	PyTorch torch.nn.functional.gelu(..., approximate='tanh')
SigmoidInfinilm(input, out)	Sigmoid(input, out)	PyTorch torch.sigmoid(input, *, out=None)

InfiniOps keeps explicit output tensors per CONTRIBUTING.md. The legacy empty GELU approximation string is normalized to PyTorch's canonical none value before delegation.