New function interface

2nd leg of kernel_callables_v3

Pullling TODO from !246 (merged) (as I don't have the permissions to edit over there.)

• Improve docs: First goals, then overall design, then details
• Avoid logic duplication map_call/map_call_with_kwargs
• Add a member function to the Reduction class that would resolve function.
• Remove function_manglers
• Fix Fixme's
• A new Program object to contain the two, where I imagine each kernel will carry a link to such a program object, and all kernels that call each other must share one. Since most people are still going to think about (and create) individual kernels first, it should get created implicitly.

Note: This does not contain CallableKernels.