This would cause us to keep track of a reference of all metrics on the
class level.

Which would mean we would keep looking at the old metrics the registry
in that we kept track of in [Prometheus::Client::Registry#metrics][0]
after it was reset using[Gitlab::Metrics.reset_registry][1].

This wouldn't cause any issues in production, because the metrics are
never reset. But in specs, we do count on those being reset in between
runs.

This is safe to do, because the registry itself is memoized on
[Gitlab::Metrics][2].

Getting the specific metric is an [initialization the first time
around][3], but after a registry is memoized, it's just a [read from a
hash][4].

[0] https://gitlab.com/gitlab-org/prometheus-client-mmap/-/blob/9e5b696dfd47a96cf22808cbebb04f00cd3d4ffc/lib/prometheus/client/registry.rb#L17
[1] https://gitlab.com/gitlab-org/gitlab/blob/82db58bab85531e49fa7264d6461928d49425a6e/lib/gitlab/metrics/prometheus.rb#L30
[2] https://gitlab.com/gitlab-org/gitlab/blob/82db58bab85531e49fa7264d6461928d49425a6e/lib/gitlab/metrics/prometheus.rb#L39
[3] https://gitlab.com/gitlab-org/gitlab/blob/82db58bab85531e49fa7264d6461928d49425a6e/lib/gitlab/metrics/prometheus.rb#L87
[4] https://gitlab.com/gitlab-org/prometheus-client-mmap/-/blob/9e5b696dfd47a96cf22808cbebb04f00cd3d4ffc/lib/prometheus/client/registry.rb#L56