TRACE improvements

[hannorein]

While working on the barycentric integrator, I'm stumbling across a few TRACE things which could be improved. I put them here for reference.

• Right now, TRACE enters an infinite loop if BS fails due to a NaN occurring at some point. This could be fixed by adding a condition like r->status <= 0 to this while loop.
• Sometimes the BS step fails for ok reasons (initial timestep too large, etc) and needs to be redone. However, even if a BS step fails, we still do a collision search. I don't think this makes sense. See this if statement.

[tigerchenlu98]

Agreed about the second point -- would it be as simple as moving the block dealing with collisions (lines 476 - 507) inside the if statement?

For the first point, what happens if BS returns a NaN? We'd just fall out of the loop and have an incomplete timestep, right? Should we return an error?