Perf/message queue

[dirkmc]

Fixes #306

go test -run=XXX -benchtime=20s -bench=. -cpuprofile prof.out

Creating a new message

Old:

         .          .    493:	// Create a new message
         .      740ms    494:	msg := bsmsg.New(false)

New:

         .          .    410:	// After processing the message, clear out its fields to save memory
         .          .    411:	defer mq.msg.Reset(false)

Adding an entry to the message

Old:

         .      480ms    518:		msgSize += msg.AddEntry(e.Cid, e.Priority, wantType, false)

New:

         .      260ms    528:		msgSize += mq.msg.AddEntry(e.Cid, e.Priority, wantType, false)

Getting the wantlist

Old:

         .      1.39s    551:		for _, e := range msg.Wantlist() {

Seems a little faster, but I think things may just have moved around. In any case we're calling it once now instead of twice:
New:

         .      220ms    417:	wantlist := message.Wantlist()

[Stebalien]

#308 may be a better solution to the CID size issues.

[Stebalien]

So, the race is actually in the tests. It looks like the virtual network hangs on to the message (and doesn't even bother serializing). We can probably just round-trip through a protobuf message with a bit of work.

[dirkmc]

It looks like the races are only showing up in the tests, I will go through and fix them

[dirkmc]

I updated the perf timings in the initial post at the top of this PR

[Stebalien]

LGTM!