Handle sqrt & Some Bodged Solution for computing through abs

[michael-schwarz]

This contains @stilscher's useful changes for computing sqrts for floats, plus some bodged together solution for refining through abs library functions we need for SV-COMP.

I propose to merge this for now and then replace it with @stilscher's cleaner solution once she finishes it.

References #1244 (and duplicates the working part of #1253).

[sim642]

Conflicts need to be resolved, but otherwise I suppose this seems ready.

[stilscher]

Based on the latest run, I wanted to take another look at some test cases that use imaxabs. We cannot handle them yet, but simply adding imaxabs to the library functions does also not seem to work since there is a debug warning about an invariant failing.

[sim642]

Since intmax_t is determined by some headers, maybe we cannot so statically know its ikind (if we can at all). It may be different from long long: https://stackoverflow.com/questions/20459513/is-intmax-t-the-same-as-long-long-int.