collector: return empty array for wildcard lookup until bg thread fil…

…ls it

The downside is that for the first ~60s wildcard metrics will not be
returned. In exchange, calls to /metrics endpoint will not be waiting for a
potentially quite long operation.

collector: don't call thread.start() twice on error in bg thread

v0.0.1-loc2

- look up keys for custom redis metrics in background thread, because Redis's
  SCAN command may exceed Prometheus scrape timeout

v0.0.1-loc1

- ResqueCollector: add a way to specify custom metrics to expose