NaN, INF read/write bug fixed

[ThakeeNathees]

Fix: #40589

writing and parsing -nan(ind) and -inf damage the .tscn file as the first negative eaten as float (-0.0) and the reset identifier is unexpected and parsing(reading) these special case would break stuff so I've changed them to nan, "inf_neg" respectively.

(regardless of the sign of the NAN, godot using -nan(ind) for both)

[bruvzg]

rtoss uses sprintf, and it can have more output variants depending on system / C runtime:

• C99 standard CRTs should output inf, nan, nan(snan), nan(ind).
• Old Visual Studio CRTs (MinGW may use it too) will output it as 1.#INF random-digits, .#IND random-digits or .#NAN random-digits.

And any of these can be prefixed by -.

[ThakeeNathees]

Now I'm checking before stringifying the the value, but not sure could it be there any negative nan
@bruvzg

[bruvzg]

I do not think there's any meaning to the sign of NaN if it is set, it's just artifact of previous operation. isnan should handle any NaN correctly and there's not reason to save it.

[aaronfranke]

@ThakeeNathees Is this still desired? If so, it needs to be rebased and tested on the latest master branch.

[briansemrau]

This is something I'd like to see merged in in 3.x as well. This issue is currently annoying me in my own project. If necessary, I'm willing to fork this branch and open new PRs.

[Calinou]

@briansemrau Since ThakeeNathees appears to be inactive, I'd suggest opening a rebased version of this pull request against the master branch, then open another pull request against the 3.x branch.

[akien-mga]

Superseded by #47497.