Fix #236.

fix bug with local_gpu_shape. Their is GPUShape that inherit from Shape and it should return the good value.

cast to/from floatX that are not needed as they don't change the type will not generate an OP. Added test for this.

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?

    InversePermutation
    ReorderRowElements

argument "ndim_added" to random_function.
Use that mechanism for "multinomial" and "permutation" distributions.

distribution.

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.