Fix race on peer state gather

[hannahhoward]

Goals

Fix final race condition on peer stats gathering. Note that there's already a lock on all the methods of PeerTaskQueue itself, this just wraps one more level so that we can hold the lock a bit longer when gathering peer state

[rvagg]

Looks like a layer of protection, but I don't know how safe range rm.inProgressRequestStatuses is without a lock - if it's possible for a range query like this to hit a synchronization error then I guess this is needed - I just don't know Go well enough to know this.

[hannahhoward]

@rvagg the issue is actually making sure no changes happen to the peer task queue while reading the status.

the simplified version of the issue goes like this:

PeerState method (inside main request manager go routine):

1. statuses = gather statuses
2. queueState = get queuestate

worker queue:

1. PopTasks()
   -----*** at this moment, queue state is out of sync with request state ***---
2. ExecuteTask(task) {
   send to request manager which will resynhronizes state
   }

The issue here is PopTasks could be called between step 1 and step 2 of PeerState

[rvagg]

oh, of course, even better. I didn't notice the fromPeerTopics call in there but that solves the full sync problem quite nicely.