lots of numerics improvements

[oscardssmith]

the main ones are sqrt(sqrt(x)) instead of x^.25 and cispi(x) instead of exp(im*pi*x). I also started converting abstoabs2` which is noticeably faster for complex, but I think I've missed a few.

[gbaraldi]

I believe modulo isn't equivalent to mod but to rem

[stla]

Yes, modulo is rem, thanks.

I've manually done the changes, because of conflicts.

I don't believe in the accuracy of cispi: I had to remove it, because one of my unit tests, a comparison with an exact value, is failing with cispi.

[stla]

Look:

julia> cispi(1im)
0.04321391826377052 + 0.0im

julia> exp(-pi)
0.04321391826377226

The second one is closer to Wolfram:

0.04321391826377224977...

[oscardssmith]

Oh, that's surprisingly inaccurate That shouldn't be happening.

[simonp0420]

The inaccuracy in cispi has been fixed, along with a 2x speedup in this PR