fix: wait until we finish connecting before we cancel the context

This is an interesting bug because changes to the DHT have suddenly started
triggering it. I'm not sure _why_ we weren't hitting it before now.

We may have been ignoring a context somewhere?

providerquerymanager/providerquerymanager.go

@@ -252,8 +252,8 @@ func (pqm *ProviderQueryManager) findProviderWorker() {
 					}
 				}(p)
 			}
-			cancel()
 			wg.Wait()
+			cancel()
 			select {
 			case pqm.providerQueryMessages <- &finishedProviderQueryMessage{
 				k: k,