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.