Merge branch 'gizmo_fix' into 'master'

Repaired GIZMO_SPH after the time integration changes that had been made.

The most important change is the addition of a new method, `hydro_timestep_extra`, that is called after the new time step for a particle has been computed, and that passes the physical time step for the next step to the particle.
In `GIZMO_SPH`, this method is used to
 - store the physical time step in `p->force.dt`. This value is used during the force loop to exchange fluxes between particles
 - reset a new `p->force.active` flag to an inactive value (`0`). This flag is set to an active value (`1`) in `hydro_init_part`, and is used during the force loop to decide whether or not to exchange flux with particle `j` in the asymmetric version of the interaction method.

For all other hydro schemes, this method is empty.

Other changes:
 - made sure `benchmarkInteractions` compiles with `GIZMO_SPH`
 - added a compilation flag, `GIZMO_FIX_PARTICLES`, that disables particle movement. If possible, this should be replaced by a corresponding parameter
 - added some sanity checks on density and pressure values, and a proper treatment of vacuum in the gradient prediction step

This version successfully runs most of the hydro examples (including the Sod tests, Sedov blasts, Kelvin-Helmholtz test, square tests, Gresho vortex), but fails to run the 2D and 3D Noh problem (1D works).

See merge request !315