[doc] Tensor.mean: dtype kwarg is not documented

[ssnl]

The following works:

>>> torch.__version__
'1.4.0a0+5635a72'

>>> x.mean(dtype=torch.float)
tensor(1.5000)
>>> help(x.mean)

>>> x.mean(dim=1, dtype=torch.float)
tensor([1.6667, 1.3333])

Yet the doc doesn't say anything about the dtype argument:

Help on built-in function mean:

mean(...) method of torch.Tensor instance
    mean(dim=None, keepdim=False) -> Tensor or (Tensor, Tensor)

    See :func:`torch.mean`

[vincentqb]

Please feel free to submit a pull request for this :)