We may end up loading lots of nested stacks concurrently with
convergence and those would try to discover endpoints and fetch
access info objects from keystone. This at times results in
ConnectTimeout errors from keystone. We can avoid these errors
by adding some retries.
Also adds retries to client_plugin get_endpoint() calls, which has
simillar issue.
Change-Id: I18cde971248eff5783f97c9e7a60316d7dd93431
Task: 36349
(cherry picked from commit 6fb8ac250a)