Retry when connection_closed is received during a streamed response
The changes_reader uses a streamed response. During the stream, it's
possible to receive a connection_closed error due to timeouts or
network issues. We simply retry the request because for streamed
responses a connection must be established first in order for the
stream to begin. So if the network is truly down, the initial request
will fail and the code path will go through the normal retry clause
which decrements the number of retries. This way we won't be stuck
forever if it's an actual network issue.
BugzId: 70400
COUCHDB-3010
3 files changed
