Support for NaN and -0

[sffc]

Related to #50, but on the separate issue of supporting NaN and -0, the other special values that are allowed in Number and IntlMV.

Many of the same arguments in favor of Infinity apply to NaN and -0: most everything else supports them, and if Amount is trying to build on top of existing numeric types, it should support them, too.

The main difference with Infinity is that the use cases are more limited. NaN is mainly useful as the result of an expression like 0/0, but we don't directly support arithmetic in Amount. I'm not aware of strong use cases for NaN in finance of scientific computing; it usually indicates a bug in the software. I feel similarly with -0.

CC @waldemarhorwat

[jessealama]

I find it hard to accept NaN and -0 when thinking about Amount. The best argument I can see is alignment with Intl mathematical values.

[jessealama]

I'd be in favor of supporting -0 as an argument (whether as a Number or as a String) but silently normalizing that to just (unsigned) zero.