rack.attack.match_discriminator is only return on
throttle_authenticated_api or throttle_authenticated_web requests, so
we're avoiding logging user_id on blacklist requests

Follow up of https://gitlab.com/gitlab-org/gitlab-ce/issues/62756

Current code uses module attribute which stores value
global instead of locally (thread locally). This results
in concurrent accesses to overwrite the each other values

Signed-off-by: Rémy Coutable <remy@rymai.me>

Use separate Prometheus metrics dirs in dev/test

Closes #63869

See merge request gitlab-org/gitlab-ce!30253

Fetches user based on the value of 'rack.attack.match_discriminator'

Closes https://gitlab.com/gitlab-org/gitlab-ce/issues/62756

Using `on_master_start` assures that the sampler is started in
master process and not in worker processes.

Store Sidekiq and Web server metrics from Prometheus in separate
directories in `development` and `test` environments.

- Add two new ActiveRecord models:
  - RootNamespaceStoragestatistics will persist root namespace statistics
  - NamespaceAggregationSchedule will save information when a new update
to the namespace statistics needs to be scheduled
- Inject into UpdateProjectStatistics concern a new callback that will
call an async job to insert a new row onto NamespaceAggregationSchedule
table
- When a new row is inserted a new job is scheduled. This job will
update call an specific service to update the statistics and after that
it will delete thee aggregated scheduled row
- The RefresherServices makes heavy use of arel to build composable
queries to update Namespace::RootStorageStatistics attributes.
- Add an extra worker to traverse pending rows on
NAmespace::AggregationSchedule table and schedule a worker for each one
of this rows.
- Add an extra worker to traverse pending rows on
NAmespace::AggregationSchedule table and schedule a worker for each one
of this rows

Loading `ApplicationSetting` from Redis was responsible for at least 50%
of the CPU load of the Redis cluster on GitLab.com. Since these values
generally don't change very much, we can load this from the database and
cache it in memory, skipping Redis altogther. We use
`ActiveSupport::Cache::MemoryStore` as a drop-in replacement for
`RedisCacheStore` even though we probably don't need synchronized access
within `Thread.current`.

Closes https://gitlab.com/gitlab-org/gitlab-ce/issues/63977

We want to call master_restart callback on start too - this callback
is used for cleaning up prometheus files. It has been added to
Omnibus already:
https://gitlab.com/gitlab-org/omnibus-gitlab/merge_requests/3350

So make sure that Unicorn example configs are up-to-date and that
the same callback is called for Puma too.