Improvement suggestion

[sguysc]

In sampling/active_intersections.py, for the function _index_active:
Is there any reason you use

idx[:] = self.lincon.integration_domain(self.ellipse.x(t + dt)) - \
                 self.lincon.integration_domain(self.ellipse.x(t - dt))

To calculate the rising and falling edges where the segments become active, rather than
something like this:

active = self.lincon.integration_domain(self.ellipse.x(t + dt))
# make it cyclic
active = np.insert(active, 0, active[-1])
idx = active[1:]-active[0:-1]

Because the integration_domain function could be quite computationally expensive
if the problem has many dimensions and it should be called as little as possible.
More explanation: since we assume linear constrained domain, if
are two adjacent intersections, then sign(ellipse.x(+dt)) = sign(ellipse.x(-dt)).