butler lacks tenacity

in #99 there was good working config for manager=prometheus and bad config for the granary manager. The bad config in granary halted all operations. Butler needs to ignore bad manager objects (maybe raise a metric like butler_config{manager=granary} = 0) and continue running the manager objects that are valid such as the prometheus object in #99

Start by reviewing how Butler loads and starts manager objects, then compare the working Prometheus configuration with the failing Granary configuration described in issue #99. Done means one invalid manager no longer halts startup or other valid managers, with any configuration metric behavior decided and covered by an appropriate test.

Written by the indexing model from the issue text.