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

Fix few tests for the nightly buildbot

Dnn pool support of pad

adding documentation to  addbroadcast() unbroadcast() and patternbroadca...

Work around cudnn bug.

Problems to fix:
- dnn pool grad with ignore_border=False do not seam to compute the good value. Did we have a bug? Do they support it? Maybe not as it isn't documented.
- We do not tests mode=average correctly as we do not have a reference implementation.
- We do not tests pad. Need a reference implementation.

Fixes and cleanup for maxpool with padding

Check negative strides in gpu_contiguous

Escape backslashes in windows path string

Fix opt crash due to new CPU parameter

support max pooling with padding

[WIP] Detect Null gradient wrt non-sequences in scan

Additional pep8 fixes

Alter test case to avoid use of advanced indexing