Some people found confusing that some functions of theano.tensor (like unbroadcast and rebroadcast) are not accessible through a method of tensor variables (say, v.unbroadcast, using self as the parameter). See* 491.

On the other hand, dimshuffle (and maybe others) is only available as a method of tensor variables, not as a function of theano.tensor.

We should either:
- decide what goes where
- mirror all these functions

I think the latter would be more appropriate.