Sprint and friends format NAN floats as 0.00

I'm handing this off to Kris, as Sprint is his territory. I think we need support for the following: +INF, -INF, NAN, +0.0, -0.0.

For what it's worth, C99 allows the following for infinity: inf, INF, infinity, INFINITY. And the following for NaN: nan, NAN.

We should probably also support both fixed and scientific notation for floating point numbers, and hex representation for floating point and integer numbers. I'm not sure how many of these are already supported, but I wanted to write this down while I was thinking about it. My old unformat code for Phobos did all this on the read-side, so it may be marginally useful as a reference for what forms of representation we might want to include (assuming the C# format spec doesn't define its own version of all this).

Float.d has always had support for these, but changed recently to try and be "type agnostic" e.g. no type-specific bit-pattern testing. The infinity test worked but the NaN test failed miserably. Should be OK now

Oh, and thanks for reporting this