High Resolution Kelvin-Helmholtz

High-resolution Kelvin-Helmholtz instabilities (2D test) do not work and crash after being unable to converge smoothing length.

Reproduction Steps

• Go to examples/HydroTests/KelvinHelmholtz_2D
• Edit the makeIC.py to ensure that there are >1000^2 particles in the run
• Run SWIFT
• Get the following output:

    1150   1.710892e-02    1.0000000    0.0000000   1.144409e-05   38   38            6            0            0            0                 6.435      0
[00198.8] runner_do_ghost: Smoothing length convergence problem: iter=21 p->id=4724349 h_init=5.93332632e-04 h_old=6.63240433e+00 h_new=6.63197374e+00 f=0.005327 f_prime=12.367292 n_sum=1.53005803e+00 n_target=1.52473104e+00 left=0.00000000e+00 right=6.63240433e+00
    1151   1.712036e-02    1.0000000    0.0000000   1.144409e-05   38   41      6290560            0            0            0              2073.738      0
    1152   1.713181e-02    1.0000000    0.0000000   1.144409e-05   38   38           17            0            0            0                 6.977      0
[00199.8] space.c:space_regrid():444: Must have at least 3 cells in each spatial dimension when periodicity is switched on.
This error is often caused by any of the followings:
 - too few particles to generate a sensible grid,
 - the initial value of 'Scheduler:max_top_level_cells' is too small,
 - the (minimal) time-step is too large leading to particles with predicted smoothing lengths too large for the box size,
 - particles with velocities so large that they move by more than two box sizes per time-step.

Code setup

./configure --with-hydro=any --with-kernel=any --with-hydro-dimension=2

Tried with intel2018 and armclang, I guess those are as far apart as you can get. Code version is the master as of this being posted.