test: add nan_to_num operator coverage

[voltjia]

Summary

• Adds handwritten nan_to_num operator coverage in tests/test_nan_to_num.py.
• Covers contiguous and strided inputs/outputs across float32, float16, and bfloat16, including NaN, positive infinity, and negative infinity replacement.

Motivation

Generated smoke coverage confirms the PyTorch backend path is wired, but it does not provide focused per-operator correctness coverage. This PR adds a dedicated nan_to_num test following the existing operator-test style.

Closes N/A

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

ruff format --check tests/test_clip.py tests/test_nan_to_num.py
2 files already formatted

ruff check tests/test_clip.py tests/test_nan_to_num.py
All checks passed!

Cross-platform cropped smoke validation:
python3 -m pytest tests/test_clip.py tests/test_nan_to_num.py -q -v --devices <cpu + platform>
72 passed

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	passed; build 149s, test 5s, total 154s	72 passed
Iluvatar	Yes	passed; build 138s, test 4s, total 142s	72 passed
MetaX	Yes	passed; build 224s, test 15s, total 239s	72 passed
Cambricon	Yes	passed; build 312s, test 8s, total 320s	72 passed
Moore	Yes	passed; build 305s, test 20s, total 325s	72 passed
Ascend	Yes	pytest passed; build 186s, test 14s, total 200s	72 passed; container exit code was 137 after pytest summary, consistent with prior Ascend runs

Full `pytest` output (optional)

NVIDIA:    tests/test_clip.py + tests/test_nan_to_num.py: 72 passed in 0.90s
Iluvatar:  tests/test_clip.py + tests/test_nan_to_num.py: 72 passed in 0.52s
MetaX:     tests/test_clip.py + tests/test_nan_to_num.py: 72 passed in 0.42s
Cambricon: tests/test_clip.py + tests/test_nan_to_num.py: 72 passed in 2.63s
Moore:     tests/test_clip.py + tests/test_nan_to_num.py: 72 passed in 13.42s
Ascend:    tests/test_clip.py + tests/test_nan_to_num.py: 72 passed in 1.72s

Benchmark / Performance Impact

N/A - test-only change.

Notes for Reviewers

This PR is split from a batch of elementwise operator tests. logit, clamp_max, clamp_min, gelu, and conj_physical are intentionally left out because they need separate backend or generated-wrapper follow-up before they can be made non-regressing across all platforms.