Fix handle_size_arg when len(size) == 0

[Pang-GJ]

修复两个问题：

flow.full

>>> import oneflow as flow
>>> flow.full([], 8)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/data/home/pangguojian/oneflow/python/oneflow/nn/modules/constant.py", line 396, in full_op
    size = _handle_size_arg(size)
  File "/data/home/pangguojian/oneflow/python/oneflow/nn/modules/utils.py", line 50, in _handle_size_arg
    assert len(size) > 0, "size of tensor doesn't exists"
AssertionError: size of tensor doesn't exists

torch中：

# torch.full
>>> import torch
>>> torch.full([],8)
tensor(8)

修改方法当len(size)==0时直接返回size

new_full

在torch中：

>>> import torch
>>> tensor = torch.ones((2,))
>>> tensor.new_full([], 2)
tensor(2.)

oneflow中：

>>> import oneflow as flow
>>> tensor = flow.ones((2,))
>>> tensor.new_full([], 8)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/data/home/pangguojian/oneflow/python/oneflow/framework/tensor.py", line 210, in _new_full
    return flow.new_full(
  File "/data/home/pangguojian/oneflow/python/oneflow/nn/modules/constant.py", line 531, in new_full_op
    size = _handle_size_arg(size)
  File "/data/home/pangguojian/oneflow/python/oneflow/nn/modules/utils.py", line 50, in _handle_size_arg
    assert len(size) > 0, "size of tensor doesn't exists"
AssertionError: size of tensor doesn't exists

应用问题一的修改后，问题变为：

>>> import oneflow as flow
>>> tensor = flow.ones((2,))
>>> tensor.new_full([], 2)
tensor([2., 2.], dtype=oneflow.float32)

这是由于_handle_meta_args中

if size is None or len(size) == 0:
        new_size = input.shape

所以返回的new_size是shape，导致了不一致

[CLAassistant]

All committers have signed the CLA.

[BBuf]

测试加一下吧

[strint]

补下测试，验证之前不能通过的 case 现在可以通过

[Pang-GJ]

done

[Pang-GJ]

测试已加