This fixes -Wfloat-equals compile errors that Alam enabled for the next branch. See https://git.magicalgirl.moe/STJr/SRB2/merge_requests/383 for that discussion.
In short, != == comparisons are invalid for floating point due to precision weirdnesses -- decimals in the 6th or 10th place may be off by 1.
So to do a not-equal comparison, you subtract the two values and see if that difference is non-zero. fabsf(X-Y) > 1.0E-36f checks for non-zero X-Y by seeing if the abs value of that difference is greater than .00000...000001 to the 36th decimal place.