Simplify computation of relevant terms in datatypes (#6885)

Simplify computation of relevant terms in datatypes (#6885)

This makes it so that we do not consider non-singleton equivalence classes of DT for non-datatype terms to be relevant. This impacts how model construction is done for datatypes.

It's not clear why this method considered such terms to be relevant, I believe this was done to mask a previous bug; regressions now pass when this method is simplified.

Doing this is definitely wrong when using central equality engine, hence the motivation for this simplification.