Should one table hold each role's matching rules? (six places decide how the family role compares today) #1260

nameparser/_match.py decides each role's behavior in six places:

• _GROUP (which vocabulary links a role reads)
• _NOT_TRIED_INTO (where an alternative reading may go)
• _ACROSS (cross-role pairs)
• the branch chain in _compare_role
• the bases= argument that turns on particles-aside and part-of-other for the family role
• the keys' hard-coded "family" / "given"

A change to one role's rule has to be made in step across them. A single per-role table (group, comparator, one-side outcome, whether an empty side conflicts, cross roles, key words) would hold it in one place.

No outcome would move: the change is a refactor, checkable with the matching oracle staying byte-identical. Raised by the simplify pass of #1221. A candidate for the 3.1 release simplify pass.