SVD Fails on Empty Tensor with Misleading cuSolver Error

[deepliao]

Issue Title: RuntimeError when performing SVD on empty tensor (shape with zero dimension)
Environment:

• Python 3.9
• MindSpore Version: 2.2.14
• Hardware: GPU

Steps to Reproduce:

import mindspore
import mindspore.ops as ops

# Create empty tensor (2 rows, 0 columns)
zero_dim_tensor = mindspore.numpy.zeros((2, 0))

# Attempt SVD decomposition
ops.svd(zero_dim_tensor)  # Throws RuntimeError

Observed Behavior:

[ERROR] DEVICE(...): WorkerJoin failed: 
----------------------------------------------------
- cusolver Error:
----------------------------------------------------
cusolver svd fail | Error Number: 3
----------------------------------------------------
- C++ Call Stack:
----------------------------------------------------
mindspore/ccsrc/plugin/device/gpu/kernel/math/svd_gpu_kernel.cc:142 RunSvd