general question (not for this PR): shouldn't this use a comparison with some margin of tolerance? or would that introduce a bug where it might miss collisions?

```suggestion float jump_height = (jumpspeed - 0.5f * gravity * peak_dtime) * peak_dtime; // s = vt - 1/2 gt^2 ```

Should I perhaps not apply the fix to the `old_move` code?

So it's just replacing Euler with Runge-Kutta...? It's definitely going to be more accurate. We should make this change!

Could be optimized to `if (newpos_f == *pos_f)`