Mainly:
- improved GpuSum.__str__
- fixed problem in make_node of IncSubtensor and Subtensor
- added PyErr_Format()s to failure-handling code of GpuSum

But it depends on some very sketchy code in cuda_ndarray, so it will need
further work.

added the pickling of the profile mode. added option to lower the relative comparaison to make a test passs. Disable check_isfinite in one case and print the diff sumary.

changed the number of iteration of test_lenet_* to have a minimum of .1s of GPU time (to have more stable speed up number) and a maximum of 10s of cpu time(to don't be too long).

fix ignore_error=True by using the requested batch_size and print the time executed and print the result at each iter.