updated documentation related to verify_grad and TensorType

Very confused why this optimization is necessary...

split the test test_GpuCrossentropySoftmax1HotWithBiasDx in two to don't use GpuDot22 in it to make it fast enought for older gpu card. So the test don't fail on older gpu card.

enabled test_mlp again as now it work correctly. Disable it when the optimization are not the normal one as it is too slow.

remove duplicate check that make exit of theano longer then necessary with compiledir with many compiled element.

Added optimization of useless elemwise node. Refactored those optimization and test to put them together.

 reverted the many copies scan used to do since it introduced bugs and the corner case it was suppose to address was fixed by Pascal L.

fix bug in conv op on the gpu that make the code crash in some case. We don't generate bad result, but read memory that was not usefull and outside the image range.