Score corruption analysis assumes constraint matches are unique

Ever since mapping and flattening were introduced to Constraint Streams, constraint matches are no longer unique. (A constraint stream may produce two distinct tuples which are equal in contents.) Unfortunately, the score analysis still assumes that every tuple has different contents.

This manifests in the following situation:

• A constraint provider includes a constraint with duplicate tuples.
• At the same time, the constraint provider triggers a score corruption in another unrelated constraint.
• FULL_ASSERT is enabled.

In this case, score corruption analysis will give a misleading message, pointing to duplicate constraint matches being inserted twice. This is a false positive, and the actual message should be pointing towards the score corruption in the second constraint.