Support calling kernel for CTarget

I'd like to automate some of the work required to call a kernel generated with the plain C target (though this might also apply to ISPC target since it can be used via C). The `__call__` method and `CompiledKernel` currently assumes OpenCL, but perhaps it could do a lookup on its `target` attribute for a `CompiledKernel` implementation which would be target specific.

The implementation would write to temp file, invoke compiler and load the resulting DLL via ctypes; the ctypes functions can be annotated from source using pycparser; though, loopy must already have that information since it generated the kernel (I just haven't looked enough to find it).

In principle, I'd also like to be able to tune code generation and compiler invocation for different profiles (OpenMP v. OpenACC; GCC vs PGI; etc).

I can do this within my project, but if it's in scope for loopy, then maybe I can contribute it.