op forwarding should wait or write locally

right now you can't read-after-write an op, since we just submit it upstream and then return, relying on the polling task to eventually (usually within 0.5s) sync the write back to our wrapped instance.

option 1: after forwarding the op upstream, wait for the poller to receive it, before returning to the client. this would be a bit fiddly.

option 2: after forwarding upstream, go ahead and just write the op to our local copy (on conflict do nothing). configure the poller to update on conflict, so the timestamp and stuff eventually syncs back to upstream but we don't reply without having it locally

it's not super clear what to do in the event of various failure combinations