feat!: align normal with PyTorch API

[voltjia]

Summary

• Expose the four PyTorch normal overload families with optional Generator.
• Add the missing scalar-mean and scalar/scalar-with-size forms.
• Keep the two previous Tensor-mean signatures as explicitly deprecated compatibility overloads.

Motivation

InfiniOps covered only Tensor mean with Tensor/scalar std and omitted generator. The public API therefore could not represent half of PyTorch's standard overload set.

Depends on #851 and #839.

API Alignment

InfiniOps API	PyTorch alignment	Reference
normal(mean: Tensor, std: double, generator, out)	Tensor mean, scalar std overload	PyTorch source
normal(mean: Tensor, std: Tensor, generator, out)	Tensor mean, Tensor std overload	PyTorch source
normal(mean: double, std: Tensor, generator, out)	Scalar mean, Tensor std overload	PyTorch source
normal(mean: double, std: double, size, generator, out)	Scalar mean/std with explicit size overload	PyTorch source

The explicit C++ output remains trailing in every overload.

Type of Change

• feat
• fix
• perf
• refactor
• test
• docs
• build / ci
• chore
• 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

accelerator-dev/nvidia:latest, CPU + NVIDIA + Torch targeted build
Targeted integration suite: 241 passed, 8 warnings in 4.22s

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	targeted wheel build passed	all four overload families passed on CPU and CUDA
Iluvatar	No	N/A	N/A
MetaX	No	N/A	N/A
Cambricon	No	N/A	N/A
Moore	No	N/A	N/A
Ascend	No	N/A	N/A

Benchmark / Performance Impact

N/A.

Notes for Reviewers

These are exactly the four PyTorch overload families identified in the audit. The former two signatures remain available with [[deprecated]].

[voltjia]

Deferred from the current interface-alignment batch. This change combines Tensor/scalar normal overloads with Generator support; both categories are intentionally postponed for a focused follow-up.