An upcoming update to rubocop-gitlab-security added additional
violations.

- To represent an authorization scope, such as `api` or `read_user`
- This is a better abstraction than the hash we were previously using.

- Use a struct for scopes, so we can call `scope.if` instead of `scope[:if]`

- Refactor the "remove scopes whose :if condition returns false" logic to use a
  `select` rather than a `reject`.

1. Get the spec for `lib/gitlab/auth.rb` passing.

  - Make the `request` argument to `AccessTokenValidationService` optional -
  `auth.rb` doesn't need to pass in a request.

  - Pass in scopes in the format `[{ name: 'api' }]` rather than `['api']`, which
  is what `AccessTokenValidationService` now expects.

2. Get the spec for `API::V3::Users` passing

2. Get the spec for `AccessTokenValidationService` passing

- They are not included automatically since `API::Users` does not inherit from
  `API::API`, as I initially assumed.

- Scopes declared in `API::API` are considered global (to the API), and need to
  be included in all cases.

- Scope declarations of the form:

    allow_access_with_scope :read_user, if: -> (request) { request.get? }

  will only apply for `GET` requests

- Add a negative test to a `POST` endpoint in the `users` API to test this. Also
  test for this case in the `AccessTokenValidationService` unit tests.

- Declaring an endpoint's scopes in a `before` block has proved to be
  unreliable. For example, if we're accessing the `API::Users` endpoint - code
  in a `before` block in `API::API` wouldn't be able to see the scopes set in
  `API::Users` since the `API::API` `before` block runs first.

- This commit moves these declarations to the class level, since they don't need
  to change once set.

- Previously, AccessTokenValidationService was a module, and all its  public
methods accepted a token. It makes sense to convert it to a class which accepts
a token during initialization.

- Also rename the `sufficient_scope?` method to `include_any_scope?`

- Based on feedback from @rymai

- Mainly whitespace changes.

- Require the migration adding the `scope` column to the
  `personal_access_tokens` table to have downtime, since API calls will
  fail if the new code is in place, but the migration hasn't run.

- Minor refactoring - load `@Scopes` in a `before_action`, since we're
  doing it in three different places.

- Move the `Oauth2::AccessTokenValidationService` class to
  `AccessTokenValidationService`, since it is now being used for
  personal access token validation as well.

- Each API endpoint declares the scopes it accepts (if any). Currently,
  the top level API module declares the `api` scope, and the `Users` API
  module declares the `read_user` scope (for GET requests).

- Move the `find_user_by_private_token` from the API `Helpers` module to
  the `APIGuard` module, to avoid littering `Helpers` with more
  auth-related methods to support `find_user_by_private_token`

The `guard_all!` method is never called, and `guard!` is not implemented. The
`doorkeeper_guard!` method is also never called, and is mostly the same as its
non-bang counterpart.

This reverts commit 3e991230.

# Conflicts:
#	app/models/project.rb

- Fix naming of API::CommitStatuses
- Ensure we use require_dependency instead of require
- Ensure the namespace is right in lib/api/api.rb, otherwise, we
  might require Grape::API::Helpers which defines the `#params` method.
  This is to avoid requiring a file multiple times and getting an "Already
  initialized constant" error.

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

This reverts commit 0a280158.

- So that the server doesn't have to be restarted for every change in dev.