feat(cambricon): add mul provider

[baominghelly]

Summary

• Add a native Cambricon slot 0 provider for mul with contiguous, broadcasted, and strided tensor support.
• Add reusable Cambricon task-range/stride-offset helpers and skip unsigned MLU cases that the torch_mlu tensor-cloning path cannot construct.

Motivation

The canonical mul operator does not currently have a native Cambricon implementation. This prevents InfiniOps from dispatching elementwise multiplication to slot 0 on Cambricon devices.

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

A targeted Cambricon mul build and test were run. The repository-wide smoke subset was not run.

cmake -S . -B /workspace/build/infiniops-pr-mul \
  -DAUTO_DETECT_DEVICES=OFF \
  -DWITH_CPU=ON \
  -DWITH_CAMBRICON=ON \
  -DWITH_TORCH=OFF \
  -DWITH_LINKED=OFF \
  -DGENERATE_PYTHON_BINDINGS=ON \
  -DINFINI_OPS_OPS=mul \
  -DINFINI_RT_ROOT=/workspace/install/infinirt-master-test \
  -DPython_EXECUTABLE=/torch/venv3/pytorch/bin/python3.10 \
  -Dpybind11_DIR=/workspace/python-user/lib/python3.10/site-packages/pybind11/share/cmake/pybind11 \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_INSTALL_PREFIX=/workspace/build/infiniops-pr-mul-site/infini
cmake --build /workspace/build/infiniops-pr-mul --parallel "$(nproc)"
cmake --install /workspace/build/infiniops-pr-mul

-- Wrapper op allowlist: mul
[100%] Built target ops

PYTHONPATH=/workspace/build/infiniops-pr-mul-site \
LD_LIBRARY_PATH=/workspace/build/infiniops-pr-mul-site/infini:/workspace/install/infinirt-master-test/lib \
/torch/venv3/pytorch/bin/python3.10 -m pytest \
  -q tests/test_mul.py --devices cambricon --maxfail=1

117 passed, 42 skipped in 1.10s

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	No	N/A - not affected	N/A - not affected
Iluvatar	No	N/A - not affected	N/A - not affected
MetaX	No	N/A - not affected	N/A - not affected
Cambricon	Yes	Targeted mul build and tests passed	Full suite not run
Moore	No	N/A - not affected	N/A - not affected
Ascend	No	N/A - not affected	N/A - not affected

Full `pytest` output (optional)

........................................................................ [ 45%]
............ssssssssssssssssssssssssssssssssssssssssss.................. [ 90%]
...............                                                          [100%]
117 passed, 42 skipped in 1.10s

Benchmark / Performance Impact

N/A - no performance benchmark was run.

Notes for Reviewers

• The fast path handles contiguous tensors with matching shapes; the general path supports broadcasting and non-contiguous input/output strides.
• Work is chunked through NRAM and distributed across Cambricon tasks with the shared kernel_utils.h helpers.
• Supported kernel dtypes are float16, bfloat16, float32, int16, int32, int64, uint16, uint32, and uint64.
• The uint16/uint32/uint64 MLU pytest cases are skipped because the current torch_mlu cloning path cannot construct those tensors; those kernel dtype paths were compiled but not exercised by this pytest run.
• The repository-wide smoke subset and full test suite were not run; the PR remains a draft.