percentage function should throw error if result would be NaN

[calvinjuarez]

As mentioned by @seven-phases-max on #1880 (comment), a call to percentage that would currently return NaN% should instead throw an error.

This is potentially a breaking change, since it would stop compiling things that currently compile successfully (albeit poorly).

Example:

With strict-math on, the following returns NaN%, but it should error, like round or floor do.

percentage(16/17)