Previously, if movers were lost the following would happen:

1) The user would get a warning,
2) The particle->i value would be shifted by move_p to be out of bounds
3) The particle never makes it to boundary_p in order to processes it

This either teleports the particle across the domain; or causes a
segfault in the next timestep.

This PR:

1) Adds some notes to the user about how serious that warning is
2) Undoes the shift, so instead of segfauliting the code now "holds" the
particle at the cell boundary

This breaks physics, but to a lesser extent than the previous
teleporting of a particle. It's still crucial the user uses a correct
value for nm

To recreate these issues just use a deck where the intial nm is set to 1
during define_species in the input deck