Update tests for scripts/papers/maxout

[lamblin]

There is already a test in pylearn2/models/tests/test_maxout.py, but it should be updated to:

• test the cuda-convnet layers if the GPU is recent enough,
• be automatically synchronized with the yaml files in scripts/papers/maxout if possible,
• check that even if the batch size does not divide the size of the training set, all examples are used

If possible, check if we can find a way of detecting if the size of the padding is not the right one.

See #569 for a more general ticket.