Loosen "ambiguity" type inference constraint

Yes a two-staged process, each greedy, sounds best. The first stage of rules that must be applied, and the second of those ambiguous rules which agree in degree but not primality. This two step process can be repeated until no further rules are applied. Luckily the existing infrastructure is already centered around applying sets of rules, so this is purely a higher-level scheduling abstraction.

_Originally posted by @lukem12345 in https://github.com/AlgebraicJulia/DiagrammaticEquations.jl/issues/84#issuecomment-2868106644_