implemented the new flag THEANO_FLAGS=gpuelemwise.sync that default to True. When false it won't wait for the gpu fct to finish and won't check the return status. cuda/tests pass in debug mode. Should we add a sync at the end of all the theano graph? How to do so?

added the theano dtype floatX with THEANO_FLAGS=floatX=float{32,64} to allow selection of the type at execution. Usefulle for GPU stuff and speed up some computation.

added THEANO_FLAGS=nvcc.fastmath (THEANO_FLAGS=fastmath as shortcut). Will add -use_fast_math to nvcc.

added the theano flags op.set_flops that will make convop in profile mode print the number of flops.

Changed RandomStream.perform so it casts the shape into a tuple, to work
around a bug in numpy.

put the default value of THEANO_BLAS_LDFLAGS into config.py and changed so that if empty, we will use numpy.dot(no c_code)

added disable the generation of conv_full_patch_stack_padded directly in GpuConv to test the hard coding of more parameter.

small fix. python don't seam to like that we put the default value of an argument to be the result of a fct call to another module.