Ordering of dimensions is not consistent across parts of the package

[shengjiex98]

In deviation(), the three dimensions for reachable_corners and nominal_trajectory are state variables, points, time; while in bounded_runs_iter(), the three dimensions for the return value are time, augmented state, min/max; finally, the return value of evol() has dimensions time, state variables.

Making these have the same ordering of dimensions will reduce the chance of making mistakes while passing arguments across functions, such as passing in a nominal trajectory to deviation() from evol(). Of course, we should wait until after the deadlines to make these changes.

[Ratfink]

See also #8, which could be done at the same time.

[Ratfink]

To match lsim from ControlSystems.jl, we should standardize on time being last. That has the advantage of making a single time instant be a consecutive block of memory, generally improving cache performance.

In the current work I'm doing on #8 (see the intervals branch), bounded_runs_iter now just returns a vector of IntervalBoxes. I think the order of the main arrays in the deviation function shouldn't change, and in fact I'd like to make a method of corners_from_bounds to make a vector of IntervalBoxes into such an array automatically. evol needs to change the order of its result. I'll also make it accept an IntervalBox as initial state, which would be handy for computing nominal trajectories.