feature: add flag, typing to DblTheory to match RFC-0001

[tslil-topos]

The particular interpretation of that RFC here is as follows:

• constraint: change no code outside of the theory definitions
• constraint: allow users of the DblTheory trait to depend on the type system to know whether hom_types are always present
• implementation trick: "shallow" dependant typing, fn hom_type(&self, x: Self::ObType) -> <Self::Kind as DblTheoryKind>::Wrap<Self::MorType>;
  • when DblTheoryKind is Categorical, Wrap = Id, when it's SetTheoretic, Wrap = Option

Notes:

• there was a blanket implementation of DblTheory from VDCWithComposites. This has been macro-ised as imlp_dbl_theory, relaxed from blanket to allow more flexible use
• the error messages in hom_op were slightly more informative, distinguishing self.unit_ext(y) from self.through_unit(cell, 0) failures, now these failures are not distiguished
• docstrings may need more work to adhere to the high standards we have already

Several other design approaches were weighed and dismissed:

• BaseDlbTheory with sub traits DblTheory and SetTheoreticDblTheory: introduces more code, more concepts to juggle, lots of duplication
• hom_type returns Option always, with flag on impl: no type system guarantees, adds .unwrap() boilerplate to call sites, forces us to specialise the blanket implementation anyway
• DblTheory<Categorical> and DblTheory<SetTheoretic> as separate traits: loss of ergonomics at call sites, would be two copies of every method, Rust doesn't pick for you
• UnitalDblTheory supertrait with user contracts: still not strong enough guarantee, have to update call sites

Closes #1099