Map rollback across remote ops prototype

[ChumpChief]

Putting this up as draft prototype for visibility, but I want to spend a little more time thinking about splitting pending op tracking from the pending data tracking. Also needs testing for the new scenarios and some basic cleanup.

A few known/intentional behavior changes to consider:

• I early exit for delete() when it would be a no-op for the local client, rather than sending a speculative delete op.
• I don't emit events for re-adding items in response to processing a clear while pending ops are outstanding (basically un-fixes AB 411). I should probably emit events for the deleted keys though at least - I think this would probably be the most expected behavior and was previously considered as an alternative resolution to 411.
• Now has consistent iteration ordering between clients, at least after outstanding ops settle, which was not guaranteed before.
• One of the stress tests passes now.