LayerNorm with `bias=False` or `elementwise_affine=False` fails to convert when using `transformer-engine`

[cyanic-selkie]

Bug description

During the conversion, a check must be made that bias and weight are not None.

Here is the error for bias=False:

File "/usr/local/lib/python3.10/dist-packages/lightning/fabric/plugins/precision/transformer_engine.py", line 174, in _convert_layers
    replacement.bias.data = child.bias.data.clone()
AttributeError: 'NoneType' object has no attribute 'data'

What version are you seeing the problem on?

v2.5

[KAVYANSHTYAGI]

### Bug Discussion:
The following error occurs when attempting to convert layers where bias is explicitly disabled:

AttributeError: 'NoneType' object has no attribute 'data'

This happens in _convert_layers when trying to access child.bias.data.clone() without verifying
that child.bias is not None. In PyTorch, modules like nn.Linear can be constructed with bias=False,
in which case the .bias attribute is set to None by design.

Example:
layer = nn.Linear(16, 32, bias=False)
print(layer.bias) # Output: None

During layer replacement, the logic assumes bias is always present and attempts to clone its .data.
This breaks if the original layer has bias=False, leading to the AttributeError on NoneType.