out-of-place native FFT: move buffer alloc inside timing

[bibikar]

This matches our Python behavior for these benchmarks.

[oleksandr-pavlyk]

Overall the change is good. Thanks!

However, removal of #incude print_buf.inc means that even for small inputs the computed Fourier harmonics are not going to be printed out.

If this is unintended, we should put the include in the loop, like:

     if (s == 0 && it == -1) {
           # include print_buf.inc
     } 

[oleksandr-pavlyk]

Looks good. Thank you!