These limitations are too strict, and their only purpose is to avoid code
size explosion. These restrictions seem obsolete, and the size problem
is solved in other places through cheap expansion limits.
The motivation is that the old code cannot deal with comparisons against
induction variant's increment.
I guess this is slightly broken, if rhs is constant but lhs needs 2x the SCEVCheapExpansionBudget,
we'd still want to expand. (Well, or they both in total should fit into SCEVCheapExpansionBudget).
But that is a separate issue.