refactor(cuda): adapt ConvInfinilm to Convolution

[voltjia]

Summary

• Implement the deprecated CUDA-family ConvInfinilm provider as a thin adapter over canonical Convolution.
• Remove duplicate convolution metadata allocation and kernel dispatch while preserving the legacy base interface and tests.

Motivation

ConvInfinilm represents the non-transposed, zero-output-padding subset of the PyTorch-aligned Convolution operator added in #882. This PR makes the canonical provider the single implementation while retaining the deprecated InfiniLM compatibility entry point.

This is a stacked PR based on #887 because canonical Convolution must copy current InfiniRT metadata views into owning metadata before it can be constructed.

No issue is closed by this PR.

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

NVIDIA validation used accelerator-dev/nvidia:latest on an A100 host. The tested stack was #887 followed by this PR:

cmake -S /workspace/InfiniOps -B /workspace/InfiniOps/build-adapters \
-G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_CUDA_ARCHITECTURES=80 \
-DWITH_NVIDIA=ON -DGENERATE_PYTHON_BINDINGS=ON \
-DINFINI_RT_ROOT=/opt/infinirt \
-DINFINI_OPS_OPS=convolution,conv_infinilm \
-DFETCHCONTENT_SOURCE_DIR_CUTLASS=/tmp/cutlass-src
cmake --build /workspace/InfiniOps/build-adapters --target ops --parallel 8
python3 -m pytest tests/test_convolution.py tests/test_conv_infinilm.py \
--devices nvidia -q
36 passed in 2.18s

Repository checks:

clang-format 21.1.8 --dry-run --Werror: passed
ruff 0.15.22 check: passed
ruff 0.15.22 format --check: 148 files already formatted
python -X utf8 scripts/generate_wrappers.py \
--devices nvidia iluvatar metax moore \
--ops convolution conv_infinilm --strict-ops
passed

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	focused stacked ops build passed	36 focused tests passed on A100
Iluvatar	Yes	strict wrapper generation passed	Native hardware not available; platform CI/owner validation required
MetaX	Yes	strict wrapper generation passed	Native hardware not available; platform CI/owner validation required
Cambricon	No	N/A - not affected	N/A - not affected
Moore	Yes	strict wrapper generation passed	Native hardware not available; platform CI/owner validation required
Ascend	No	N/A - not affected	N/A - not affected

Full focused `pytest` output

Running 36 items in this shard
.................................... [100%]
36 passed in 2.18s

Benchmark / Performance Impact

N/A. The adapter calls the same canonical convolution kernel and this PR makes no performance claim.

Notes for Reviewers

Alignment table

Legacy interface	Canonical interface	Open-source target
ConvInfinilm(input, weight, bias, padding, stride, dilation, groups, out)	Convolution(input, weight, bias, stride, padding, dilation, transposed=false, output_padding=zeros, groups, out)	PyTorch ATen convolution schema

• The adapter only reorders existing attributes and supplies the subset constants transposed=false and zero output_padding.
• The legacy class, public signature, backend registrations, and tests remain in place and remain deprecated.
• The adapter forwards the selected stream to the cached CudaConv<Backend, Convolution> provider before every call.
• Review fix(ops): copy convolution tensor metadata views #887 first; this PR intentionally contains no copy of its metadata-view fix.