Refactor floating point comparisons to use epsilon-based functions for improved precision

[tinboxw]

In the amd64 and arm64 platform architectures, there may be inconsistencies in floating-point comparison calculations. As shown in the following figure, it is the debugging result of the IntersectsSegment function. I took the same input and output different results under different architectures.
I used the epsilon method to solve this problem.

arm64:

amd64:

[tidwall]

Can you please provide the inputs and varying platform details?

[rickb777]

This seems like a good PR to me.

I suggest making some of the functions into methods: specifically

• func PointEqual(a, b Point) bool could be func (a Point) Equal(b Point) bool
• func RectEqual(a, b Rect) bool could be func (a Rect) Equal(b Rect) bool

and then the PointNotEqual, RectNotEqual functions are not required.

This would slim down the API footprint, which would be easier for users.