Suggestion: remove extinct species from the simulation and restart solve() on every extinction.

[iago-lito]

As discussed earlier and also there.

This alternate option to killing zombies (#65) would speed up simulation by requiring less callback work, less testing in tight dBdt! loops and less degenerated zero-biomass species towards the end of massive-extinction simulations. Also, it would enable fearless use of arbitrary solving algorithms, even implicit, LU-factorization-based ones.

Unfortunately, the user would be handed several solve() results instead of one, so we would need to stich trajectories together for better experience and make this simulation strategy optional.

Again: this is not for the current milestone, and only a plan for the future ;)