sky_imager speedups

[cbassa]

Some different implementations of the sky_imager function, using different approaches to speed up the calculations, are given here.

The speed results are as follows:

numpy_real: 6.080s
numpy: 6.484s
numpy_float32_ravel: 4.013s
numpy_float32: 4.853s
cupy: 0.281s
simple: 6.530s
numba: 3.483s

Using numpy functionality, the best combination appears to be a combination of float32, using ravel to only compute on sky pixels, computing only the real part of the complex multiplication, and using numpy broadcasting.