Make NumPy types for u0 and usol configurable

[juhannc]

First of all, awesome code! I got a speedup of an order of magnitude basically for free!

But, as far as I can tell, u0 and usol are required to be of type np.float64. However, often times np.float32 is more than enough.

Maybe I just don't see the option to change it, but if it doesn't exist, it'd be nice if we could change it.

Thank you!

[Nicholaswogan]

Thanks! Ya, it should be possible to just compile the underlying C++ code for either float64 or float32. I will leave this open and get it to eventually.