libffi benchmarking

[atgreen]

It's uncommon to do a ffi_prep_cif within an inner loop. Virtually all users run ffi_prep_cif once, cache the results, and re-use it for multiple ffi_call invocations. I don't expect libffi to be much faster, but as long as you are publishing benchmarks, you might as well use it the way people in the real world use it. Thanks!