More robust implementation of PMComplex >> sqrt

[olekscode]

The implementation of PMComplex >> sqrt was improved in PR #221.

However, this implementation still suffers from potential floating point overflow/underflow.
This is why I suggest something even more robust (and more accurate) like:

Implementing Complex Elementary Function Using Exception Handling
ACM Transactions on Mathematical Software - October 1994
Ping Tang and 3 other authors

See https://source.squeak.org/trunk/Kernel-nice.1459.diff

[SergeStinckwich]

I guess we can reuse quite easily the Squeak implementation done by @nicolas-cellier-aka-nice

[nicolas-cellier-aka-nice]

Just correct the comment: it does not pick the root with positive imaginary part, but rather the root with positive integer part and same sign bit for imaginary part...

[nicolas-cellier-aka-nice]

And also correct pure imaginary section !
See https://source.squeak.org/trunk/Kernel-nice.1460.diff