Because there will be similar observer for PumaWorkerKiller,
it makes sense to keep both on better place.

Update server API to enable switch prometheus endpoint

See merge request gitlab-org/gitlab-ce!29186

PumaWorkerKiller is used for periodically checking and killing
workers (the biggest one) if overall memory reaches specified
limit. This metric allows us to watch number of killed workers.

Extract acme double to helper

Create ACME challanges for pages domains

* Create order & challange through API
* save them to database
* request challenge validation

We're saving order and challenge as one entity,
that wouldn't be correct if we would order certificates for
several domains simultaneously, but we always order certificate
per domain

Add controller for processing acme challenges redirected from pages

Don't save acme challenge url - we don't use it

Validate acme challenge attributes

Encrypt private_key in acme orders

This allows the BetterErrors REPL to be available
when your GitLab instance is listening on
something _other_ than localhost.

As we have a central domain for auto merge process today,
we should use a single worker for any auto merge process.