and out of dimshuffle_lifter

This fixes existing failing tests

Now it should work.

Also move introduction of dimshuffle into another optimization,
and update tests

Sso it can match reshape(dimshuffle(...), ...)

Bump the max version of cudnn since we work with 5.1.

Flake8 new api

doc: Fix string Fourrier -> Fourier

Partial evaluation in CVM

Ignore new star import error F405 in flake8 2.6+, change dev doc to install flake8<3 due to API change

GpuArray BatchNorm

Add gpu_contiguous for batch normalization.

Make the old backend support as_input_f32 theano.config.dnn.precision.

Float16 convolutions