Push down negation before decompose

[IgnaceBleukx]

Re-order the transformations such that negation is pushed down into the arguments of a "not" operator before we call decompose.

This allows to implement efficient implementations of negated global constraints (e.g., ~Table becomes Negtable).
We can also do this for some more (e.g., ~AllDiff(x) becomes NValue(x) < len(x)). We should probably do some small experiments to see what is more efficient? Also depends on the decomp of NValue of course if that one is not supported...

Downside is that no transformation can explicitly introduce negation, and needs to call recurse_negation themself. This turned out to be simpler to implement than I first thought.

Efficiency-wise, there might be some gains because we now have one transformation less to be called in each recursiveflatten_constraint call. I then also noticed we call simplify_boolean in each recursive flatten call, I've added a flag to disable it when we can. Not sure if this should remain in this PR

[tias]

you will have to rebase/merge master in I guess, now that push down negation rewrite is in (in a different form)

[tias]

almost there.

I guess you checked that no global cons/func introduces a negation, even on the latest master?

Also, we since have 'scip' as a solver, can you check that all solvers on master that need changing are changed?

[IgnaceBleukx]

Processed the review, there were indeed a couple of cases where the decomposition/negation of a global might introduce negation.

Updated scip and docs

[IgnaceBleukx]

Also updated the newly added highs solver, should be ready to go in after tests complete?