refactor(cuda): adapt InfiniLM softmax provider

[voltjia]

Summary

• Replace the standalone CUDA implementation behind SoftmaxInfinilm with a thin adapter to the canonical Softmax provider introduced by feat(cuda): add canonical softmax providers #890.
• Preserve the deprecated InfiniLM schema and its existing tests; this PR changes provider ownership only.

Motivation

The deprecated compatibility operator should not retain a separate CUDA kernel after the canonical, PyTorch-aligned operator exists. Sharing the provider keeps legacy callers working while removing a duplicate implementation path.

Depends on #890.

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=softmax,softmax_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	48 passed, 36 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_softmax.py tests/test_softmax_infinilm.py -q
48 passed, 36 skipped in 3.60s

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
SoftmaxInfinilm(input, dim, dtype, out)	Softmax(input, dim, dtype, out)	PyTorch torch.nn.functional.softmax(input, dim=None, ..., dtype=None)

The canonical interface preserves the Python-level argument order (input, dim, dtype), followed by InfiniOps' explicit output tensor per CONTRIBUTING.md.