Prepare 0.7rc2

[WIP] Disable some optimization that could cause problem with cudnn v2 rc3.

Fix connection_pattern and add test

Fix the broadcastable pattern of tensordot's output

Fix in local_reduce_join with axis=None

Skip flake8 if not installed

Fix problem with border_mode = <int>

Fix infer_shape for pooling, skip unimplemented test cases

Include pep8 autocheck test + while list (from blocks) #2503

Add missing import

First attempt at fixing issue 2454: making consider_constant() functiona...

fix gh-2581

Convert inputs of maxpool to tensor

Correctly handle gradient of identity on int

Flatten C code

[BUG] pool with padding

do not try cudnn opt when it isn't available

Included changes in tensor/opt.py for #2454. These were previously lost due to problems committing to GitHub.

Set borrow=False for scan inner inputs