Optimize solve_resolution_order a bit

[nstrong-scw]

== DETAILS This doesn't change the strategy very much, but a couple optimizations:

• don't resolve entities with dependencies if there are none
• in each resolution loop, only iterate over entities that didn't resolve in the previous loop

I had difficulty running the tests. Anything that invoked alembic gets a KeyError: 'formatter' error, which seems to indicate it's not loading alembic.ini correctly, but I'm not sure how to solve that.

However, I did do a basic smoke test with my own sqlalchemy/alembic/alembic_utils project.

[olirice]

You're right that it is unnecessary to loop over entities that have already been resolved, but they are skipped immediately

The cost of checking inclusion in a list is negligible compared to

with simulate_entity(sess, entity, dependencies=resolved):

so I wouldn't expect measurable performance gains from the change and it introduces additional variables to track so I'd prefer not to add the complexity

[olirice]

if you'd like to open another PR to move the log statement I'm game for that!