Increase default absolute zero tolerance in Floats

[mforets]

No description provided.

[schillic]

Any reason?

[mforets]

Good question and the short answer is that experiments suggest that sqrt(eps(Float64)) is too small for the discrete post. A "typical" calculation with the bouncing ball and octagon directions needs ~5e-8 to take the intersection properly, and since sqrt(eps(Float64)) = 1.4901161193847656e-8 then it's not that far.

I plan to provide more details soon -- probably after we add the global way to handle the abstol, making it easier for tests in Reachability.

[schillic]

Hm, okay.