Handle non-bias convolution layers and optional symmetry breaking noise in "wider" function

[erogol]

wider function did not work for no bais layers. I handled this. And the paper says that a small symmetry breaking noise is necessary for wider operation. I implemented this as well. Any comment would be great.

[soumith]

hi eren, there seems to be a change in either the spacing or tabs. When I look at the diff in the "Files changed" tab above, there are many lines that are not changed but show up as changed. Any chance you can fix this?