lock/query: make robust against paddles errors

lock/query: make robust against paddles errors

Retry paddles requests, and for get_status() return an empty dict
rather than None so callers behave.

get_status() failing in particular has caused the dispatcher and jobs
to fail several times over the past few weeks. With this change, we
should be able to run multiple paddles workers again, since all the
common callers will retry on error.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>