Clipping fix to underflow.

[Averydx]

Fixes the associated issue with probability of movement underflowing in mm.centroids. Probability of movement is now clamped to a minimum of $e^{-100} \approx 2.68 \times 10^{-43}$. This is close enough to zero that it shouldn't affect any simulation results. We should put a warning in the docs that mm.centroids is not well suited to the state level spatial scale.

[Averydx]

It may be better to throw an error here, except there could be cases where some nodes are geographically close and others are very far away. Then this is the preferred solution, as movement being close to zero is intended behavior. Think mainland US vs Hawaii.

[JavadocMD]

Alternatively we could make it a Centroids argument for how to deal with errors, with clipping as the default behavior. Not entirely dissimilar from the way you can specify how the CDC ADRIOs recover when there's missing or sentinel data.

rume = SingleStrataRUME.build(
  ...
  mm=mm.Centroids(fix_underflow="clip"), # default
  ...
)

The other option that comes to mind being "raise" for raising an error if it happens.

Suggest bringing this up with the group for comment.

[JavadocMD]

It would be really nice to have a battery of tests for Centroids as well. I don't think we have tests for any of the MMs yet... not easy to test stochastic things; but we might be able to test the dispersal_kernel property easily enough.