2.0.2

fix(api): reuse a service's container row when an install retries

A service's container row is inserted before its image is pulled. When the
pull failed, the job retried, found no recorded service to reuse, inserted the
same install-<id>-<service> name again and died on the unique index, so every
failed install reported 'duplicate key' instead of its real error. The row is
now looked up by name when the install has no record of it, and it takes the
current image and network, so a compose edit that changes an image applies on
redeploy instead of keeping the first one.