Conditional mul_add chunk gen speed up.

[kralverde]

@julianbraha I saw you made the original PR for using mul_add to speed up float operations in chunk gen. Those we ultimately reverted because that function was extremely slow on non-FMA cpus.

I was wondering if you had access to a FMA cpu and could run benchmarks to see if it would be worth adding a compile time method to use mul_add or x * a + b based on whether the CPU supports.

[vyPal]

Maybe something like this could be used: https://crates.io/crates/multiversion

[kralverde]

A simple cfg check should suffice for this