fix!: align embedding input order with PyTorch

[voltjia]

Summary

• Restore the public Embedding input order to input, weight, matching torch.nn.functional.embedding.
• Propagate the corrected order through the shared CUDA and Ascend implementations and the existing embedding tests.

Motivation

PR #799 accidentally changed the leading tensors from the public PyTorch order to the lower-level ATen order weight, indices. These are the most important positional arguments, and accepting them in reverse order makes the user-facing API misleading and incompatible with torch.nn.functional.embedding.

This is a follow-up to #799; no standalone issue is associated with the correction.

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

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

Host: ssh nvidia
Image: accelerator-dev/nvidia:latest
PyTorch: 2.10.0a0+b4e4ee8

cmake ... -DINFINI_OPS_OPS=embedding
cmake --build build-nvidia --target ops -j 16
Build passed.

python -m pytest tests/test_embedding.py --devices cpu cuda -q
42 passed

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	embedding-target build passed	tests/test_embedding.py: 42 passed
Iluvatar	Yes	Not run locally - no Iluvatar host available	Uses the changed shared CUDA implementation
MetaX	Yes	Not run locally - no MetaX host available	Uses the changed shared CUDA implementation
Cambricon	No	N/A - not affected	N/A - not affected
Moore	Yes	Not run locally - no Moore host available	Uses the changed shared CUDA implementation
Ascend	Yes	Not run locally - no Ascend host available	Source order updated consistently

Full pytest output

Running 42 items in this shard
..........................................                               [100%]
42 passed

Benchmark / Performance Impact

N/A. This changes argument order and names only; the kernel implementation is unchanged.

Notes for Reviewers

API alignment

Interface	Signature	Alignment basis
InfiniOps primary interface	embedding(input, weight, padding_idx, scale_grad_by_freq, sparse, out)	PyTorch inputs in Python order; explicit out remains last per CONTRIBUTING.md.
PyTorch public API	F.embedding(input, weight, padding_idx=None, max_norm=None, norm_type=2.0, scale_grad_by_freq=False, sparse=False)	Pinned PyTorch source

• The short overload is also restored to embedding(input, weight, out).
• max_norm and norm_type remain out of scope because PyTorch handles renormalization in its Python functional wrapper before calling the low-level embedding operation.
• The incorrect feat!: align non-InfiniLM operator interfaces #799 order cannot be retained as a deprecated overload: both orders have the same C++ type sequence, so parameter names cannot distinguish overload resolution. This PR restores the pre-feat!: align non-InfiniLM operator interfaces #799 public order instead of adding runtime dtype-based ambiguity.
• clang-format 21.1.8 passed with --dry-run --Werror; Ruff passed for tests/test_embedding.py.