Improve robustness of sampling test for harmonic oscillator

[ijpulidos]

From time to time we get stochastic test failures for the TestHarmonicOscillatorsMultiStateSampler with the following traceback:

Traceback (most recent call last):
  File "/home/runner/micromamba-root/envs/openmmtools-test/lib/python3.10/site-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/home/runner/work/openmmtools/openmmtools/openmmtools/tests/test_sampling.py", line 230, in test_with_unsampled_states
    self.run(include_unsampled_states=True)
  File "/home/runner/work/openmmtools/openmmtools/openmmtools/tests/test_sampling.py", line 223, in run
    raise Exception("Dimensionless free energy difference exceeds MAX_SIGMA of %.1f" % MAX_SIGMA)
Exception: Dimensionless free energy difference exceeds MAX_SIGMA of 6.0

Even if these are commonly worked around just by re-running the workflow we probably need a more robust way of testing this system.