Update Cython usage in scan_perform.pyx

[brandonwillard]

We're currently versioning the C code produced by Cython for the Scan Op, and I'm not sure if that's necessary (or the correct approach). It also requires a manual compilation and patch step when the relevant code is changed, which seems a little less than desirable.

Since it looks like we're compiling the Cython-generated C code with Theano's GCC compilation framework, I'm guessing that Cython's compilation framework somehow isn't/wasn't compatible. Even so, this situation raises some important questions:

1. If we can—apparently—use Cython to generate C code that's compatible with Theano's C framework, why aren't we doing that more often? Better yet, why don't we have a Linker that does this automatically, and why don't we use something like that to replace all these string-based c_code implementations?

Cython already has a well developed means of both separating and combining Python code from its more implementation specific C details, so using that framework could dramatically improve the unnecessary tangle of high and low-level details brought about by the current Theano C transpilation framework.

2. Why can't we use Cython's compilation framework?

There's really no reason to think that Cython's compilation framework isn't already far ahead of Theano's in many ways (e.g. default parameters and host system detection/configuration for one). My initial guess is that we would need to use it programatically, at least in a way that's similar to what Theano is currently doing with gcc and the like, but I don't see any reason why it wouldn't be just as amenable—if not considerably more (e.g. if it has a Python interface that helps us avoid writing our own onerous system calls).

[brandonwillard]

Closed by #963