Fix some issues that may cause catastrophic floating point error

[user202729]

Made the following changes

• change p+d*t to q+d*(t-1). The problem is if -a+sqrt(det) ≫ 1, then the term arg(v, q) is supposed to be identically zero, however if q ≈ (0, 0) then it's possible that p+d ≉ q and arg(v, q) ≉ 0, which results in a large error. See a.cpp in the attached file below.
• even if side != (s.cross(e, prev) < 0) is true, it is possible that the 4 points are almost collinear that lineInter returns the second value being (0, 0). With this change, whenever (a < 0) != (b < 0) then 0 ≤ a / (a - b) ≤ 1, which means the point being pushed is always on the line between prev and cur i.e. no catastrophic error can happen. See b.cpp in the attached file below.

Attachment (demonstration of catastrophic failure): a.zip

[simonlindholm]

These are fantastic catches. Thank you so much!