Timestep criterion for zoom in simulations

This is an issue found by Yves in Gadget-2's timestep criterion.

When running our simulations with Pop III stars, we end up with particles that have extremely low mass (the question of if we should run with such low mass particles is something that we should discussed within our group, but it does not matter for this issue) and our simulations slow down artificially due to the global time step restriction given here https://gitlab.cosma.dur.ac.uk/swift/swiftsim/-/blob/master/src/engine.c#L3030. This is based on the lowest mean distance between particles, it means that for massive particles, we will impose a far lower timestep than required:

     /* Inter-particle sepration for the baryons */
     const float d_b = cbrtf(min_mass_b / (Ob * rho_crit0));

In the case of zoom in simulations, we should maybe change it in order to impose the timestep individually (or locally) to each particle according to their mass and not the minimal mass of the simulations.

This condition is slowing us down by a factor of 5 when running with Pop III stars even if the galaxies (at all redshift) are comparable.

What do you think @matthieu?