Flatten tags in graphite output

[eberkut]

When using the influx output, tags are "flattened" but this is not the case for the graphite output which can create duplicate keys and make the graphite output unusable.

For instance, for a micrometer Java application:

1. Raw Prometheus output
   logback_events_total{level="error",} 0.0
   logback_events_total{level="warn",} 0.0
   logback_events_total{level="info",} 123.0
   logback_events_total{level="debug",} 0.0
   logback_events_total{level="trace",} 0.0

2. sensu prometheus collector influx output
   logback_events_total,level=error value=0 1546955856
   logback_events_total,level=warn value=0 1546955856
   logback_events_total,level=info value=124 1546955856
   logback_events_total,level=debug value=2 1546955856
   logback_events_total,level=trace value=0 1546955856

3. sensu prometheus collector graphite output
   logback_events_total 0 1546954471
   logback_events_total 0 1546954471
   logback_events_total 124 1546954471
   logback_events_total 2 1546954471
   logback_events_total 0 1546954471

To cope with Graphite grammar for metric names compared to influx, multiple tags could be concatenated with '.' (instead of ',' for influx). In addition, '/' could be converted to '_' and '=' to '.'.