Low Level Non Allocating API

It would be great to have non allocating variants of the FFT functions.

```julia
vA = rand(ComplexF64, 1000);
@allocated fft!(vA)
```

I get some allocations (Using `FFTW.jl`).
It would be great to have a low level API which has no allocations at all.
It might require an auxiliary function to generate the needed workspace.

It will be appreciated in the context of static binaries generated by Julia.