Replace job:up base metric with the health check's probe_success

Look at healthcheck prober metrics to assess target state, rather than
using Prometheus' own "up" metric. This allows us to take advantage of
redundant blackbox probers by removing noise caused by individual
prober failures.

To do so without changing the semantics, we add a new 'float_job'
label on the health check probes, that maps to the Prometheus target names.