It seems to me that we're currently just picking the first CN codec, rather than the one that matches the clock rate of the voice codec. The only test I've gotten to fail by changing this behavior is the one that's also changed in this CL, which explicitly expects a CN codec to be chosen even though there's none matching.
BUG=webrtc:7282
Review-Url: https://codereview.webrtc.org/2707133007
Cr-Commit-Position: refs/heads/master@{#16979}