fix(ascend): correct rotary embedding handling

[baominghelly]

Summary

• Correct the Ascend RoPE implementation by using separate aclnnRotaryPositionEmbedding calls for query and optional key tensors.
• Extend the Ascend RoPE test coverage for the supported forward full-dimension NeoX configuration.

Motivation

The previous Ascend RoPE implementation used an ACLNN layout and semantic combination that produced incorrect query and key values. It also required a key tensor even though the operator interface permits query-only execution.

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

$ cmake -S /workspace/InfiniOps \
    -B /workspace/InfiniOps/build-rope-pr952 \
    -DCMAKE_BUILD_TYPE=Release \
    -DCMAKE_INSTALL_PREFIX=/workspace/InfiniOps/test-rope-pr952/infini \
    -DGENERATE_PYTHON_BINDINGS=ON \
    -DWITH_ASCEND=ON \
    -DBUILD_ASCEND_CUSTOM=ON \
    -DSOC_VERSION=Ascend910_9382 \
    -DINFINI_RT_ROOT=/workspace/InfiniRT-6b256e0-prefix \
    -DINFINI_OPS_OPS=rotary_embedding
$ cmake --build /workspace/InfiniOps/build-rope-pr952 -j8
[ 55%] Built target infiniops
[100%] Built target ops

$ ASCEND_LAUNCH_BLOCKING=1 python -S -m pytest \
    tests/test_rotary_embedding.py --devices ascend -q -rs --tb=short -x
6 passed, 18 skipped, 14 warnings in 1.28s

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	No	N/A - not affected	N/A - not affected
Moore	No	N/A - not affected	N/A - not affected
Ascend	Yes	build passed	RoPE targeted suite passed: 6 passed, 18 skipped; full suite not rerun

Full `pytest` output

.sss.sss.sss.sss.sss.sss

SKIPPED [18] tests/test_rotary_embedding.py:45:
Ascend supports forward full-dimension NeoX RoPE

6 passed, 18 skipped, 14 warnings in 1.28s

Benchmark / Performance Impact

N/A. This is a correctness fix; performance was not benchmarked.

Notes for Reviewers

• Ascend currently supports forward full-dimension NeoX RoPE; unsupported variants remain explicit test skips.
• Query and optional key tensors are processed independently through ACLNN, then copied back to preserve the operator's in-place behavior.
• Validation used InfiniRT 6b256e0ab0853d9e2a7ef059bac22c847bd02251 and SOC_VERSION=Ascend910_9382.