feat(cambricon): add CNCL backend for local multi-device collectives

[baominghelly]

Summary

Add full Cambricon CNCL backend support and integrate it with existing CCL/MPI architecture.

This PR adds CNCL device and backend integration, common CCL communicator initialization, CNCL rank-based initialization for native and MPI-hybrid multi-threaded programs, CNCL collectives and point-to-point communication, and the required build, dispatch, and example updates.

Changes

CNCL Backend Integration

• Add CNCL CMake configuration and auto-detection for Cambricon environments.
• Register CNCL in backend manifests, device/backend compatibility traits, backend priorities, and bridge generation.
• Add CNCL API wrappers for:
  • AllReduce
  • AllGather
  • Send
  • Recv
  • GetUniqueId
  • CommInitRank
  • CommDestroy
• Add CNCL error-checking macros and Cambricon-specific API bindings.
• Increase INFINICCL_UNIQUE_ID_BYTES to accommodate cnclCliqueId.

Communicator Initialization

• Implement CNCL rank initialization through cnclInitComms.
• Coordinate multiple local thread requests into a single process-local CNCL initialization call.
• Support native multi-threaded initialization and MPI-process plus multi-threaded MxN initialization.
• Validate clique IDs, global ranks, local devices, duplicate ranks, and duplicate devices.
• Queue separate initialization groups until the active CNCL group completes.
• Refine common CCL CommInitRank handling and safely copy backend-specific unique IDs.
• Update OpenMPI communicator initialization to preserve local communicator-size metadata.

Point-to-point Communication

• Add compatibility handling for CNCL versions that do not accept a nullptr queue.
• Preserve asynchronous behavior for caller-provided queues while synchronizing internally created fallback queues.

Documentation Update

• Update README.md and .github/pull_request_template.md to include CNCL as a valid supported backend.

Platform and Backend Affected

Platform

• CPU
• NVIDIA GPU
• Iluvatar GPU
• MetaX GPU
• Moore Threads GPU
• Cambricon MLU
• HYGON DCU

Backend

• OpenMPI
• MPICH
• NCCL/RCCL
• MCCL
• CNCL

OpenMPI is marked because its CommInitAll implementation is adapted to the corrected multi-handle interface; its communication path remains unchanged.

NCCL/RCCL and MCCL are marked because common ccl implementation has changes.

Performance Impact

• No performance impact
• Performance improved
• Performance regression possible

The branch enables native CNCL communication instead of relying on fallback paths. No formal benchmark comparison is included.

Known Issues & Future Work

• CNCL allows each clique to be initialized only once per process, so local initialization requests must be coordinated and serialized.
• CNCL currently does not support float64 reduction data or the Avg reduction operation.
• CNCL versions up to 1.30.8 require a non-null queue for point-to-point operations.
• CommInitAll is not supported yet.
• A future improvement could generalize communicator-role-aware backend selection for APIs that support both intra-node CCL and inter-node MPI backends.

Test Results

Test Involved Platform

• CPU
• NVIDIA GPU
• Iluvatar GPU
• MetaX GPU
• Moore Threads GPU
• Cambricon MLU
• HYGON DCU

Test Involved Backend

• OpenMPI
• MPICH
• NCCL/RCCL
• MCCL
• CNCL

Cambricon CNCL + OMPI:
ccl_mpi_hybrid_all_gather.log
ccl_mpi_hybrid_all_reduce.log
ccl_mpi_hybrid_send_recv.log

Cambricon CNCL:
ccl_all_gather.log
ccl_all_reduce.log
ccl_send_recv.log

NVIDIA NCCL + OMPI:
ccl_mpi_hybrid_all_gather.log
ccl_mpi_hybrid_all_reduce.log
ccl_mpi_hybrid_send_recv.log

---

Checklist

Title, Branch, and Commits

• PR title follows Conventional Commits.
• Branch name follows the repository branch naming convention.
• Each commit message follows Conventional Commits.
• Relative to the stacked base, this PR is a single squashable commit.
• No stray merge commits from master are present.
• No fixup!, squash!, or wip commits remain.

Scope and Design

• Changes are minimal and limited to CNCL support plus the required CommInitAll interface correction.
• No dead code, debug output, or unowned TODOs were introduced.
• No unrelated formatting churn was introduced.
• The existing public C API signatures remain unchanged.

General Code Hygiene

• Comments are limited to non-obvious intent.
• Modified and added files end with trailing newlines.
• git diff --check passes.
• Comments and error messages are in English and follow repository conventions.

C++ Specific

• Code follows the repository C++ style.
• Applicable formatting was checked on the validated implementation.
• No exceptions are introduced.
• Error handling follows repository conventions.
• N/A: No constructor initializer-list changes are included.

Python Specific

N/A: This PR does not modify Python files.

Testing

• Applicable local multi-device CNCL operations were built and tested successfully on Cambricon hardware.

Build, CI, and Tooling

• CNCL backend auto-detection and build registration are included.
• git diff --check passes; hosted CI will run on this Draft PR.

Documentation

• README backend documentation can be added following maintainer feedback.
• N/A: This change is additive and has no user-visible breaking change.

Security and Safety

• No secrets, internal URLs, customer data, or personal hardware identifiers are included.
• No third-party source code is introduced.
• Communicator inputs and CNCL type mappings are validated before dispatch.