feat(cambricon): add Embedding implementation

[baominghelly]

Summary

• Add a native Cambricon Embedding implementation with a CNNL fast path for contiguous tensors and a custom BANG C path for scalar and strided tensors.
• Support max_norm renormalization and introduce reusable CNNL/CNRT resource helpers.
• Extend embedding coverage to scalar index tensors.

Motivation

Enable native Embedding execution on Cambricon while preserving the existing InfiniOps interface for max_norm and layouts that CNNL does not accept directly.

No linked issue.

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

The required affected-platform smoke build and tests -m smoke set have not been run after the latest commit. Focused Cambricon validation was run in container 0f34305d9bc9 on cambricon218_01:

/torch/venv3/pytorch/bin/python3.10 -m pytest tests/test_embedding.py -q --devices cambricon
............................................................             [100%]
60 passed in 0.33s

A clean Cambricon infiniops/ops build and installation also completed successfully during focused validation.

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	Focused clean build and installation passed; required smoke suite not run	tests/test_embedding.py: 60 passed
Moore	No	N/A - not affected	N/A - not affected
Ascend	No	N/A - not affected	N/A - not affected

Focused validation output

clang-format version 21.1.2
clang-format --dry-run --Werror   src/native/cambricon/cnnl_utils.h   src/native/cambricon/cnrt_utils.h   src/native/cambricon/ops/embedding/kernel.h   src/native/cambricon/ops/embedding/kernel.mlu
# passed with no diagnostics

git diff --check upstream/master...HEAD
# passed with no diagnostics

/torch/venv3/pytorch/bin/python3.10 -m pytest tests/test_embedding.py -q --devices cambricon
............................................................             [100%]
60 passed in 0.33s

Benchmark / Performance Impact

N/A - this is a new backend implementation; no baseline benchmark was collected.

Notes for Reviewers

• CNNL handles contiguous, non-scalar forward calls. The custom BANG C path handles scalar or strided forward calls and max_norm renormalization.
• The custom forward path zero-fills output rows for out-of-range indices; reviewers should confirm this Cambricon behavior is acceptable because PyTorch raises IndexError.
• cnnl_utils.h and cnrt_utils.h centralize reusable CNNL descriptor/handle ownership and CNRT device-buffer ownership for future Cambricon operators.
• The launcher declaration is in kernel.h and its implementation is in kernel.mlu, matching existing Cambricon operator practice. This differs from the literal launcher-placement wording in CONTRIBUTING.md and should be confirmed by maintainers.
• Before marking this PR ready, rebase onto current master and rerun the required affected-platform smoke build and smoke test set.