ENH: We could allow converts to be written in C

[seberg]

I already abstracted conversion away into it C-functions. Right now Python converters are handled slightly differently (but only very slightly, all they have is an additional userdata field holding the python function really).

There is nothing hindering us from making the converters public API. It just needs some definition of how they should be passed. scipy.LowLevelCallable would not be a real option I guess, but basically that, just a notch simpler.

One advantage of adding something like this, is that in case we got more annoying parsing idiosyncrasies, we could "offload" easily deal with them in C, although I don't currently think we would need that.