fix(ascend): detect exact SoC version

[baominghelly]

Summary

• Prefer the exact device name returned by torch_npu, such as Ascend910_9362, when automatically selecting Ascend SOC_VERSION.
• Resolve the detector Python from _TORCH_PYTHON, Python_EXECUTABLE, or PATH in that order.
• Keep npu-smi info as a fallback, but only accept model-qualified names such as 910B4 or Ascend910_9362.
• Preserve Ascend910B4 as the final fallback for existing environments where an exact model cannot be determined.
• Add CMake script tests for exact torch_npu detection, rejection of HBM numbers such as 3109, and model-qualified npu-smi fallback.

Motivation

Ascend 910C hosts can report the generic name Ascend910 through npu-smi, while AscendC compilation requires the exact SoC name Ascend910_9362. The previous text search could also mistake an HBM value such as 3109 / 65536 for an Ascend 3109 device. This makes automatic custom-kernel builds select the wrong SoC or fall back unnecessarily.

Closes #

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
• 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

Affected-platform smoke build was not run for this final rebased head. The changed detector was validated directly with:

docker run --rm \
  --security-opt seccomp=unconfined \
  -e PYTEST_ADDOPTS="-p no:cacheprovider" \
  -v "$PWD":/src:ro -w /src \
  infiniops-ci/ascend:cann9.1.0-torch2.10.0-npu-post4-20260909 \
  python3 -m pytest tests/test_ascend_soc_detection.py -q

3 passed, 14 warnings in 0.09s

Real Ascend 910C detector result:

device: Ascend910_9362
-- DETECTED_SOC=Ascend910_9362

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	Not run - targeted detector test and real 910C detector validation passed	Full suite not run; change is limited to Ascend SoC detection

Full `pytest` output (optional)

Not run - full Ascend smoke suite is deferred; this draft only changes SoC detection and has targeted detector coverage.

Benchmark / Performance Impact

N/A

Notes for Reviewers

• An explicitly supplied SOC_VERSION still takes precedence over automatic detection.
• torch_npu is preferred because it returns the exact runtime device model.
• The fallback deliberately rejects bare Ascend910/Ascend310 and numeric-only tokens to avoid interpreting HBM values as SoC names.
• Real-hardware validation was performed on Ascend 910C (Ascend910_9362). The existing fallback preserves 910B4 behavior, but 310P was not hardware-validated for this change.