Gpu op for solve using cusolver (CCW #5191)

Fix issue #4476 related to BatchedDot.

Force install of the new commit of Theano

Otherwise, it is just using the cached one...

Use GpuAdvancedIncSubtensor1_dev20 more frequently in the new back-end

Extend new tests.

Add related test in theano/tensor/tests/test_basic.py

Change C code cache version.

Remove opt warning from local_dimshuffle_subtensor

Merge CumsumOp/CumprodOp into CumOp

Make GpuAdvancedIncSubtensor1_dev20 being introduced for vector_scalar case instead of GpuAdvancedIncSubtensor1

Make GpuAdvancedIncSubtensor1_dev20 support mixed dtype for x and y. Most of the code was already supporting this.

Fix tests when floatX=float32.

Update checking of integer types. 

in a `grad()` function.

Add Theano flag cmodule.age_thresh_use

This fixes issue #5274.
For integer types, any string-based type-checking is replaced with a list-based type-checking using the type lists provided in `theano.tensor` (int_dtypes, uint_dtypes, integer_dtypes, etc.).

NB:
* Module `theano.sparse` have its own type lists, so they are used in this module instead of `theano.tensor` lists.
* I have tried to check the types with `discrete_dtypes` where it seems relevant (e.g. in `grad()` functions), but further analyzes may be required. That is the type list to use to fix issue #5274.

fix small python 3 compat issue

but why travis can't detect this?

Handle zero-length dimensions in GpuDnnSoftmax

Minor inconsistency in AbstractConv_gradInput implementations

Fix a test in DebugMode

Use a default value compatible with broadcastable of type

Remove printing in test.