GitLab

This applies to EE queries only. It improves performance of those
queries significantly.

Related to https://gitlab.com/gitlab-org/gitlab-exporter/issues/58

Remove obsolete database probes that have been moved to the
`postgres_exporter`.
* `blocked_queries`
* `slow_queries:`
* `stuck_idle_in_transactions`
* `vacuum_queries`

In the previous commit we're using a wildcard to select all columns,
now by using PGs information_schema we query if the column exits.

The query we used refferenced a EE column directly, breaking CE compat.
This query changes that behaviour by using a wildcard match on all
projects colums.

The code which interprets the results can already handle extra columns.
Query was copy-pasted from
https://gitlab.com/gitlab-org/gitlab-monitor/issues/25#note_30414761
and solves gitlab-org/gitlab-monitor#25

* Update metric name to match new convention.
* Remove unecessary labels.
* Update specs.

Use `namespace_id` in place of `namespaces.path` in order to obfuscate
the namespace in the metrics.

* Don't create the "other" buckets until after all other buckets are
  done.

* Combine all running workers into one query
* Update specs.

* Remove `_total` from metric name, this is a gauge.
* Split `running` and `created` builds into separate metrics.
* Add `namespace.path` as new metric label.
* Refactor `metrics_total` function to support new metrics and labeling.
* Reduce cardinality for `pending` and `created` builds by ignoring
  values less than 10.
* Split stale builds metrics to separate function.
* Update specs.