Title: DNS timeouts cause no healthy upstream

[optional Relevant Links:]
I have a test case outlined here on how to reproduce this. I have verified this happens on master branch too.

Bug Template

Description:
In the event that DNS resolver has transient errors (i.e timeout) envoy currently doesn't check the c-ares status of SUCCESS before overriding the address_list with an empty array. We have observed in production cases where we have healthy hosts, and the next time a DNS query happens if it were to timeout, our service ends up going down.

Repro steps:
Test case to reproduce. a simple iptables block on the DNS resolver will do.

https://github.com/chrisgoffinet/envoy-dns
Patch: chrisgoffinet@950c734