Allow distributed wells with partition method zoltanwell

[michal-toth]

The parameter --allow-distributed-wells was till this commit ignored, it was always set to false. With these changes, the value true makes the partitioner completely ignore wells, the partitions are calculated as if there were no wells. It is possible to obtain partitioning that has one or more wells split over multiple subdomains.

Internally, setting --allow-distributed-wells=true makes the partitioner skip the vertex contraction step in the graph representation of the grid, where cells of each well were merged to one (which prevented the partitioning algorithm from splitting any well).

Documentation changes: Partitioner zoltanwell (default) no longer ignores the allow-distributed-wells. This makes its behavior more similar to other partitioners.

[lisajulia]

jenkins build this please

[lisajulia]

@michal-toth: I've tested this with NORNE and it looks similar (unfortunately not quite as good as) what I have with #824. Thanks!
Can you please:

• rebase this on master (I tried and got some failures)
• clean the commit history (i.e. remove merge commits and commits where conflicts are resolved)
  Thanks!

[atgeirr]

This PR does not require any code changes to work with OPM/opm-simulators#6264 if that is merged, but it will change testing (and also the title of this PR would be wrong...) so be aware.

[michal-toth]

jenkins build this serial please

[lisajulia]

@michal-toth: I would love to merge this, but can you clean the commit history before, i.e. rebase this once more and remove merge commits? Thanks!

[michal-toth]

@michal-toth: I would love to merge this, but can you clean the commit history before, i.e. rebase this once more and remove merge commits? Thanks!

The merge commits are gone. However, commit messages got messed up, I will look up how to amend them.

[akva2]

git rebase -i origin/master, replace 'pick' with 'r(eword)'.

[lisajulia]

Thanks!

[lisajulia]

jenkins build this serial please