Mutation simulations are an integral part of Django Evolution. They're

responsible for validating that a mutation can be performed against part

of a database signature, and modifying that signature for storage. The

way they worked before is that each mutation had a simulate() method

that took in several pieces of state, and it was the responsibility of

the method to look up data (and handle errors during lookup) from the

signatures and, if necessary, raise a SimulationFailure if the

simulation could not be applied.
This change formalizes some of this process and simplifies the work

required in the simulate() methods. There's now a top-level

Simulation class, which olds all the state for a simulation, provides

helpful accessors for getting parts of the signature (raising errors if

not found), and providing a way to fail a simulation.
Each mutation now has a standard error message describing a simulation

failure for that mutation, which is called when failing the simulation.

This can reference variables on the class or from calls to

Simulation.fail.
Each mutation's simulate() method takes only a Simulation and no

other state, and can use that to look up the signatures. Since they no

longer have to handle signature-related lookup failures, each mutation

is now a bit smaller than it was when the improved error messages were

introduced.
Speaking of error messages, now that we have a standard way of failing a

simulation, the remaining error messages now provide detailed context,

helping to debug problems when a mutation fails.
In order to benefit SQLMutation, its update_func (which is needed to

do a simulation and update the signature) can now take either a legacy

app_label, project_sig argument, or a new-style Simulation. Newer

evolution files should use the latter.

Unit tests pass.