In a replicated setup, when an object is updated in one cluster but
still waiting to be replicated to the other cluster, GET requests with
if-match, and range headers will likely fail. It is better to proxy
requests instead.

Also, this commit avoids printing verbose logs about precondition &
range errors.