Add Cudnn version to Theano cache key to the handle_type and _RNNSplitParams. Theano was trying to reload wrong cmodule and crash. For order of code, I needed to move much code around, but that moved coded didn't changed

Wrap Op params for theano.tensor.nnet.corr3d.BaseCorr3dMM and subclasses.

Add support for cudnn v6 dilated convolution.

 - direction (enum list)
 - dH, dW, dD: int64
 - dilH, dilW, dilD: int64
 - padH, padW, padD: int64

This influences all sub-classes:
 - theano.tensor.nnet.corr3d.Corr3dMM
 - theano.tensor.nnet.corr3d.Corr3dMM_gradWeights
 - theano.tensor.nnet.corr3d.Corr3dMM_gradInputs

Test for fix in libgpuarray, plus fix in batched_dot opt

Fix jenkins tests

Give a good error when people use device=gpu

Update nnet.txt

Fixes mismatch between the formula in implemented function in the docs

Partially factorize GpuKernelBase.get_params() and configure it to use ParamsType.

remove theano/compat/six.py

Allow c_support_code() to support list of strings and update ParamsType

Closes: #5761

Update ParamsType.c_code_cache_version().

(small update) Add restrictions to warning messages condition in GpuDnnConvGradW

 - Show warnings only if floatX==float32.
 - Show warnings only for cuDNN prior to v6.

add gpuarray to python2 buildbot

Mention duplicated values in warning messages.

Adapt local_gpu_elemwise optimization of new gpuarray back-end to avoid number of inputs overflow with Elemwise<add,mul>. The current optimization was already splitting the input, but it was not using the method split_huge_add_or_mul because of new gpuarray lifter signature (see comment 
https://github.com/Theano/Theano/pull/5852#discussion_r114145523)

The unit test for large number of inputs was invalid because it was testing the old back-end (theano.sandbox.cuda). It is now adapted to gpuarray lifter optimization function. The number of settings tested is reduced to lower the computation time while still making sure we test at least one case with no number of inputs overflow and at least one case with number of inputs overflow.

split_huge_add_or_mul() is made more general so it can be used if any case like Elemwise<add,mul> occurs elsewhere.

BUG: fix bytes/string concat error, closes #5887