This refactors the TodosFinder finder according to the new code reuse
rules, as enforced by the CodeReuse cops. I also changed some of the
methods to use regular if statements, instead of assignments and/or
early returns. This results in a more natural flow when reading the
code, and it makes it harder to accidentally return the wrong result.

In EE it will render a Geo Status widget when Geo is enabled and it is
in a secondary node.

Also added minimal specs to that action.

Auto DevOps deployment strategies now supports timed incremental
rollout. We are deprecating the usage of INCREMENTAL_ROLLOUT_ENABLED
environment variable in Auto DevOps template.

The new behavior will be driven by the INCREMENTAL_ROLLOUT_MODE variable
that can either be manual (same as INCREMENTAL_ROLLOUT_ENABLED) or
timed.

Rollout deployments will be executed using a 5 minute delay between each
job.

Adds CI_SERVER_VERSION_MAJOR, CI_SERVER_VERSION_MINOR,
CI_SERVER_VERSION_PATCH to the list of environment variables
passed to CI jobs.

EE needs :request_store to cache license queries - otherwise we will
blow past the query limit just with `SELECT * FROM licenses` over and
over.

We don't think this is needed any more - see
https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/21521, and
https://gitlab.com/gitlab-org/gitlab-ce/issues/52271 for removing the flag.

Rails5 has params[:file_type] as '' if file_type is included as nil
in the request

This will allow an explicitly-disabled feature flag to override a
feature being available for a project.

As an extreme example, we could quickly disable issues across all
projects at runtime by running `Feature.disable(:issues)`.

Moves the related_branches method from Issue model
to RelatedBranchesService

Changes `Banzai::CrossProjectReference#parent_from_ref` to return the
project in the context if the project's `full_path` matches the ref
we're looking for, as it makes no sense to go to the database to find a
Project we already have loaded.

If we don't call #to_a, we're relying on the members already being loaded from
elsewhere. Otherwise we'll do a separate query for each user:

    [1] pry(main)> Project.first.team.members.include?(User.first)
      Project Load (0.7ms)  SELECT  "projects".* FROM "projects"  ORDER BY "projects"."id" ASC LIMIT 1
      ↳ (pry):3
      User Load (1.8ms)  SELECT  "users".* FROM "users"  ORDER BY "users"."id" ASC LIMIT 1
      ↳ (pry):3
      User Exists (0.6ms)  SELECT  1 AS one FROM "users" INNER JOIN "project_authorizations" ON "users"."id" = "project_authorizations"."user_id" WHERE "project_authorizations"."project_id" = $1 AND "users"."id" = $2 LIMIT 1  [["project_id", 1], ["id", 1]]
      ↳ (pry):3
    => true
    [2] pry(main)> Project.first.team.members.to_a.include?(User.first)
      Project Load (12.8ms)  SELECT  "projects".* FROM "projects"  ORDER BY "projects"."id" ASC LIMIT 1
      ↳ (pry):1
      User Load (9.6ms)  SELECT "users".* FROM "users" INNER JOIN "project_authorizations" ON "users"."id" = "project_authorizations"."user_id" WHERE "project_authorizations"."project_id" = $1  [["project_id", 1]]
      ↳ (pry):1
      User Load (0.6ms)  SELECT  "users".* FROM "users"  ORDER BY "users"."id" ASC LIMIT 1
      ↳ (pry):1
    => true

These can be triggered by project creation in the setup phase if a spec uses the
RequestStore, but we really don't care about that - it's not an N+1, it's just
several projects being created!