Update regression case for workforce scheduling

[g-poveda]

• we identified instance_170 as one "unlucky" instance where the cpsat auto was regressing. Though, through deeper and larger studies (https://github.com/g-poveda/discrete-optimization/blob/gpd/investigate_reg/examples/workforce/scheduling/investigate_reg.py) we are confident that it's actually not a regression of the cpsat auto.

[nhuet]

There are still regressions but it is improving :)

• ubuntu: no more regressions!
• macos: regressions on instance 147: 6, 150 instead of 4, 12 ...
• windows: not poping since Improve non-regression tests handling #532 not yet merged, but
  • instance 147: 5,67 au lieu de 4, 12
  • instance 252: 10,26 au lieu de 9, 84