scan/scan_op: Convert known_grads to OrderedDict

jenkins test print commands

More fixes for Jenkins

Pre cleanup

Remove useless reshape

Add device in the compiledir_format

Fix #4809

This was probably causing a different order of operation
during gradient computation in scan for each run. With this
fix I'm able to finally reproduce results on my RNN system.

Fix issue #4809

Handle filter_dilation in the abstractconv_grad* perform methods

Fix jenkins

We have to make the distinction between Op types in the original and
optimized graph

That way it can work with @parameterized and GPU

The tolerances are just a starting point, we may need to adjust them
later.

jenkins: explicit python paths and group all gpu tests together

Add tensor5, dtensor5 etc.

* Installation documentation refactor

* Remove the extra install links from the homepage

* Restore pics

* Rename css extension to suppress warning

* Fix install is not a toctree so is not indexed

* Fix multiple labels with same name

* Fix stdint.h link

* Remove reference to old GCC download instructions

* Fix colors

* Fix installation instructions

* Fix troubleshooting

* Fix broken link in dev_start_guide

* Prevent doc generation script from using GPUs

* Add numpy and scipy max version

* Remove Canopy references

* Fix script fails if THEANO_FLAGS is not set

* Add warning for deprecation of the old backend + minor changes

* Default env_th_flags should be an empty string

* Remove PYTHONPATH reference from doc, add libgpuarray options first

* Add static link to the dev installation instructions

* Fix syntax of link

Allow indexing using Ellipsis.

sandbox/cuda: Fix crossentropy typos