Expose floating point environment and handle fp exceptions

[boggle]

libmath provides a rich interface to deal with numerical exceptions, rounding modes, and the like.

We need to figure, if and how much of it we want to (e.g. as a resource that clears on destruction)

Useful references and any comments on this very welcome

Additionally:

• decide if and when to fail in the presence of unhandled floating point exceptions
• verify sane concurrency behavior
• beyond complex numbers that is the last part of libmath that needs wrapping

[catamorphism]

Not a backwards-compatibility thing. Nominating for milestone 3, feature-complete.

[graydon]

accepted for feature-complete milestone

[bblum]

Does libmath use global state for tracking anything?

[catamorphism]

Not 1.0

[thestinger]

Closing in favour of #10186.