feat(cambricon): add fill provider

[baominghelly]

Summary

• Add a native Cambricon slot 0 provider for fill in src/native/cambricon/ops/fill/.
• Support host scalar and scalar Tensor values, contiguous and strided outputs, in-place execution, and empty tensors.
• Use a dedicated raw 64-bit path for int64 and uint64, and skip unsupported Cambricon float64 test cases.

Motivation

InfiniOps did not have a native Cambricon provider for the canonical fill operator. This prevented Cambricon builds that explicitly select implementation 0 from executing fill operations through InfiniOps.

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 operator build and test were run after rebasing onto the latest master. The repository-wide smoke build and smoke test set were not run, so this PR remains Draft.

cmake -S . -B /workspace/build/rebase-fill-20260901 \
  -DAUTO_DETECT_DEVICES=OFF \
  -DAUTO_DETECT_BACKENDS=OFF \
  -DWITH_CPU=ON \
  -DWITH_CAMBRICON=ON \
  -DWITH_TORCH=OFF \
  -DWITH_LINKED=OFF \
  -DGENERATE_OPERATOR_CALL_INSTANTIATIONS=ON \
  -DGENERATE_PYTHON_BINDINGS=ON \
  -DINFINI_OPS_OPS=fill \
  -DINFINI_RT_ROOT=/workspace/install/infinirt-master-test \
  -DPython_EXECUTABLE=/torch/venv3/pytorch/bin/python3.10 \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_INSTALL_PREFIX=/workspace/build/rebase-fill-20260901-site/infini
cmake --build /workspace/build/rebase-fill-20260901 --parallel "$(nproc)"
cmake --install /workspace/build/rebase-fill-20260901
PYTHONNOUSERSITE=1 \
PYTHONPATH=/workspace/build/rebase-fill-20260901-site \
LD_LIBRARY_PATH=/workspace/build/rebase-fill-20260901-site/infini:/workspace/install/infinirt-master-test/lib:/usr/local/neuware/lib64:/usr/local/neuware/lib \
/torch/venv3/pytorch/bin/python3.10 -m pytest -q \
  tests/test_fill.py --devices cambricon --maxfail=1

[100%] Built target ops
48 passed, 6 skipped in 0.34s

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 fill build passed	tests/test_fill.py: 48 passed, 6 skipped; repository smoke/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)

..............................ssssss..................                   [100%]
48 passed, 6 skipped in 0.34s

Benchmark / Performance Impact

N/A — no performance benchmark was run.

Notes for Reviewers

• FillUnion handles FP16, BF16, FP32, and integer types up to 32 bits using NRAM-sized chunks.
• FillRaw64Union handles int64 and uint64 by copying the exact 8-byte value, avoiding unreliable typed 64-bit scalar handling in the MLU kernel path.
• Contiguous outputs use bulk NRAM-to-GDRAM copies. Strided outputs use the shared logical-index-to-offset helper added by the merged Cambricon mul provider.
• Cambricon float64 is intentionally not registered and its six test cases are skipped. Unsigned 16/32/64-bit paths are compiled but are not exercised by the current torch_mlu pytest input set.
• clang-format --dry-run --Werror and git diff --check origin/master...HEAD passed. The repository-wide smoke and full suites were not run.