Prevent VM_rint outputting negative zero

[k9ur]

As the name suggests, VM_rint rounding to an int should output a value that can actually be represented by an int, which negative zero can't.
Previously values from -0.5 to 0.0, including +0, -0, and not -0.5, would round to -0. This meant rounding a value known to be non-negative could output negative zero. This function now outputs +0 for the aforementioned values.

If for whatever reason it's actually intentional that it can sometimes output -0, I'd suggest changing the diff to be just the following (below), so that at least VM_rint(0.0) == 0.0, rather than VM_rint(0.0) == -0.0. It'd then be functionally identical to C's round function as far as I can tell.

-	if (f > 0)
+	if (f >= 0)

Haven't tested the code, unable to, reviewer please do so.

[terencehill]

Slightly more optimized implementation:

	if (f >= 0.5) // >= 0 works too but it would be slightly less optimized 
		PRVM_G_FLOAT(OFS_RETURN) = floor(f + 0.5);
	else if (f > -0.5)
		PRVM_G_FLOAT(OFS_RETURN) = 0; // prevents negative 0
	else
		PRVM_G_FLOAT(OFS_RETURN) = ceil(f - 0.5);

[terencehill]

Btw, for consistency's sake VM_ceil should be patched too as it returns -0 if input is > -1 and < 0

[k9ur]

Btw, for consistency's sake VM_ceil should be patched too as it returns -0 if input is > -1 and < 0

I'm not sure if I agree with that about VM_ceil. To me, it'd make more sense if VM_ceil and VM_floor behaved the same way as C's math.h ceil and floor respectively, and ceil does output -0.
Meanwhile VM_rint isn't named VM_round so I think it's fine to have it behave differently to C's round.

Thanks for the optimization, I'll change the diff to that later, after a maintainer responds to my comment:

If for whatever reason it's actually intentional that it can sometimes output -0, I'd suggest changing the diff to be just the following (below), so that at least VM_rint(0.0) == 0.0, rather than VM_rint(0.0) == -0.0