Improve adaptive time step

[MaxThevenet]

In the current adaptive time step, Rank r computes step s and calculates (from beam properties at step s) the time step it will use to go from step s+nranks to s+nranks+1. At step s, rank r does not know what the physical time will be at step s+nranks (it will receive when starting step s+nranks), it just calculates the dt it will use then. An easy improvement would be to use the current Ez (at step s) to better estimate what will be the energy at the beginning of step s+nranks. A rough estimate would be min(uz-Ez*nranks*dt). This will not be too accurate because the time step is adaptive, so a better estimate would be obtained by also including adaptive time steps (just analytically calculate how much deceleration occurred after nranks/nt_per_betatron betatron periods or so). Note that, as we do a prediction of the momentum at s+nranks, we should also do a prediction of the position, to evaluate better the density and avoid artifacts, see #884. A soon-to-be-open PR will also store the min_uz (over beams and slices etc.). For now it will contain the min_uz at step s, but should also be updated to contain min_uz at step s+nranks.

[MaxThevenet]

Done in #893