StackLight's data model needs improvements

[Patrick Petit]

The data model used for metrics (Heka message structure) are somewhat misleading and incomplete.

Several improvements are needed.

1 - The 'Logger' field should contain the name of the (telemetry) plugin that emitted the metric (ex. collectd_cpu)
2 - The 'source' field should be removed from the data model because it overlaps with the Logger field and is of no actual use.
3 - An 'entity' field should be created to identify the resource or service the metric applies to (ex.: disk, cpu, nova-api, cinder-scheduler, rabbitmq, ...)
4 - The 'tag_fields' field should be renamed 'dimensions' to explicitly express the fact that it references a list of fields representing dimensions of the metric that can be referenced in an alarm rule.

The documentation (user and developer) must be updated accordingly.

In particular, the documentation should state that the fields referenced in 'dimensions' can be used in the alarm rules and are stored as metadata 'tags' of the metric in the time-series database (InfluxDB).