Simplify witness invariants for bitfield domain

[sim642]

I just happened to look into a witness we now produce with the portfolio at a level which enables the bitfield domain and saw lots of trivial invariants like (i & 0) == 0. The non-relational witness invariant simplification (#1517) happened before the bitfield domain was introduced. The latter wasn't added to witness invariant tests, so this went unnoticed.

This is a bit unfortunate because we now have a bunch of logic for simplifying non-relational and relational witness invariants, but then we also added these, so the SV-COMP 2026 witnesses might again be unnecessarily large...