There is a version of this I have in mind that would take at least 6-8 weeks of pretty solid developer time. A new layer would be built on top of SQLAlchemy ORM that would look just like Django mappings/QuerySet. The mappings wouldn't be too difficult, QuerySet maybe a little more.
The advantage to that approach would be that existing Django mappings and QuerySet code could be reused in the context of a SQLAlchemy session conversation, that is, the unit of work/identity map paradigms would be brought in. A lot of the "bolted-on/doesnt-really-work" code I see to approximate these features in a tool like JohnnyCache wouldn't be needed anymore.
I'm not sure that this really encompasses "manage" though, I'd have to increase my familiarity with that tool to know everything it does.
Complexity breeds bugs like rabbits; if you can't look at something and understand what's going on (potentially drilling down into an abstraction or two), then it's going to be a maintenance nightmare.
If you can come up with something that's easy to use and powerful, like the declarative_base stuff in SQLAlchemy over the top of the Mapper/Relationship, then I'm all ears - but if complexity is the price that I have to pay for that then it's not an easy tradeoff to make, and in a lot of cases you're better off with the easier, worser solution.
A quick question: do you have any idea of how much work it would take to port SQLAlchemy to Django properly? ie. so that all of the fancy Django manage commands work. I'm assuming a fair bit, but it might be a good way to get some sort of help as far as SQLAlchemy goes. Or you might be happy to continue forging ahead solo, I don't know. Just a thought ;)