reimplement fmaf

[japaric]

to not use any f64 operation.

Rationale: the current implementation makes use of f64 and it has bad performance on architectures that have no hardware support for double precision floats.

Both the MUSL and newlib implementations use f64 ops internally. We'll need to find some other implementation or come up with one ourselves.

What needs to be done:

Move the existing MUSL port into a private math::musl module and make sure we continue compiling it.
Write the new implementation into src/math/$fun.rs
Write a new test generator for this function.

[tgross35]

As an update, I am adding generic implementations which will avoid using the same-sized float.