Timesteps sometimes take negative time.

When we were obtaining performance results with SWIFT here, we noticed on the Xeon processors certain timesteps were outputting negative time taken:

 131   1.023438e-03   7.812500e-06     262144          0               502.015
 132   1.031250e-03   7.812500e-06     262144          0               508.461
 133   1.039063e-03   7.812500e-06     262144          0               503.477
 134   1.046875e-03   7.812500e-06     262144          0               506.939
 135   1.054688e-03   7.812500e-06     262144          0               501.405
 136   1.062500e-03   7.812500e-06     262144          0             **-1184.532**
 137   1.070313e-03   7.812500e-06     262144          0               511.000

I was running the 3D sedov blast with 3 threads (for the above result) yml file below. Is this a known issue @pdraper @nnrw56 @matthieu ?

# Define the system of units to use internally.
InternalUnitSystem:
  UnitMass_in_cgs:     1   # Grams
  UnitLength_in_cgs:   1   # Centimeters
  UnitVelocity_in_cgs: 1   # Centimeters per second
  UnitCurrent_in_cgs:  1   # Amperes
  UnitTemp_in_cgs:     1   # Kelvin

# Parameters governing the time integration
TimeIntegration:
  time_begin: 0.    # The starting time of the simulation (in internal units).
  time_end:   2e-3  # The end time of the simulation (in internal units).
  dt_min:     1e-5  # The minimal time-step size of the simulation (in internal units).
  dt_max:     1e-5  # The maximal time-step size of the simulation (in internal units).

# Parameters governing the snapshots
Snapshots:
  basename:            sedov # Common part of the name of output files
  time_first:          0.    # Time of the first output (in internal units)
  delta_time:          1e-2  # Time difference between consecutive outputs (in internal units)

# Parameters governing the conserved quantities statistics
Statistics:
  delta_time:          1e-3 # Time between statistics output

# Parameters for the hydrodynamics scheme
SPH:
  resolution_eta:        1.2348   # Target smoothing length in units of the mean inter-particle separation (1.2348 == 48Ngbs with the cubic spline kernel).
  delta_neighbours:      0.1      # The tolerance for the targetted number of neighbours.
  CFL_condition:         0.1      # Courant-Friedrich-Levy condition for time integration.

# Parameters related to the initial conditions
InitialConditions:
  file_name:  ./sedov.hdf5          # The file to read