feat(nccl): add all-gather communication

[voltjia]

Summary

Implement the existing infinicclAllGather API for the NCCL backend, including communicator-aware backend selection, zero-count handling, and reliable NVIDIA and mixed-backend validation.

This PR is stacked on #58, which provides the shared supported-backend selection used by collective and point-to-point operations. #58 is stacked on #57, which keeps generated communication bridges current when provider headers change.

Changes

• NCCL AllGather
  • Map the generic AllGather operation directly to ncclAllGather while preserving NCCL's per-rank count and legal in-place buffer contract.
  • Register the NCCL provider and validate the communicator backend, device, and handle before dispatch.
• Backend selection
  • Reuse communicator-aware selection to fall back to a supported intra-node or inter-node backend without dispatching unsupported combinations.
  • Return success for zero-count operations, infinicclNotSupported when the communicator has no compatible provider, and an internal error when it has no initialized backend.
• Validation
  • Add a two-GPU example covering out-of-place and NCCL-defined in-place AllGather with element-wise result validation.
  • Extend the backend-selection CTest for unsupported communicators, OpenMPI fallback, no-backend, and zero-count behavior.
  • Make the existing MPI AllGather example propagate validation failures through its exit status.

Platform and Backend Affected

Platform

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

Backend

• OpenMPI
• MPICH
• NCCL
• MCCL

Performance Impact

• No performance impact
• Performance improved
• Performance regression possible

N/A. This PR enables an existing API for NCCL and makes no performance claim.

Known Issues & Future Work

• Other CCL backends remain unchanged and continue to report their existing support status.
• Numerical integration coverage in this PR uses float32; datatype acceptance remains governed by the existing backend type map.

Test Results

Environment: ssh nvidia, accelerator-dev/nvidia:latest, two NVIDIA A100-SXM4-80GB GPUs, clean Release builds.

• clang-format 16.0.6 --dry-run --Werror passed for every changed C++ file.
• A clean [cpu,nvidia] / [nccl] full build passed, followed by CTest 2/2.
• CUDA_VISIBLE_DEVICES=0,1 examples/ccl/all_gather passed out-of-place and NCCL-defined in-place element-wise validation and printed AllGather validation passed.
• A clean [cpu,nvidia] / [ompi,nccl] full build passed, followed by CTest 2/2.
• mpirun --allow-run-as-root --bind-to none -np 2 examples/mpi/all_gather passed with Correct: YES and rank-ordered blocks [r0: 1] [r1: 2].
• git diff --check passed.

Test Involved Platform

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

Test Involved Backend

• OpenMPI
• MPICH
• NCCL
• MCCL

---

Checklist

Title, Branch, and Commits

• PR title follows Conventional Commits.
• Branch name follows <type>/xxx-yyyy-zzzz where <type> matches the PR title's Conventional Commits type and words are joined with hyphens.
• Each commit message follows Conventional Commits.
• This small PR is a single squashable commit.
• No stray merge commits are present; this PR is cleanly stacked on feat(nccl): add point-to-point communication #58.
• No fixup! / squash! / wip commits remain.

Scope and Design

• Changes are minimal and limited to NCCL AllGather and its reliable validation.
• No dead code, commented-out blocks, debug prints, or unowned TODOs were introduced.
• No unrelated formatting churn was introduced.
• The existing public C API signature is unchanged; this PR provides its NCCL implementation.

General Code Hygiene

• Comments are limited to non-obvious intent.
• Every modified and added file ends with a trailing newline.
• No trailing whitespace, inconsistent indentation, or mixed formatting styles remain.
• Identifiers in comments and error messages use Markdown syntax where applicable.
• Comments and error messages are in English.
• Comments and error messages follow repository sentence conventions.

C++ Specific

• Code follows the Google C++ Style Guide and the repository .clang-format.
• clang-format version 16.0.6 passed with --dry-run --Werror.
• No exceptions were introduced.
• Error and warning wording follows repository conventions.
• No constructor initializer-order issue was introduced.
• Required blank-line conventions are satisfied.

Python Specific

N/A. This PR does not change Python files.

Testing

• All applicable AllGather examples were built and run successfully on NVIDIA with NCCL and with the mixed OpenMPI/NCCL configuration described above.

Build, CI, and Tooling

N/A. This PR does not add a backend or device to auto-detection. The repository's C++ formatting gate was reproduced locally; hosted CI will run on the pushed branch.

Documentation

• The new validated usage is documented by examples/ccl/all_gather.cc and this PR description.
• No user-visible breaking change is introduced.

Security and Safety

• No secrets, access tokens, internal URLs, customer data, or personal hardware identifiers were committed.
• No third-party source code was added.
• Buffer capacities and the NCCL-defined in-place offset are validated by the two-GPU example; no uninitialized reads were introduced.