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

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

`.init()` was added in
https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/10858

We need to add it the MR diff instance that was added in a separate MR
and of course didn't add a conflict because it is a completely different
piece

The problem is that we often go via a diff object constructed from the diffs
stored in the DB. Those diffs, by definition, don't overflow, so we don't have
access to the 'correct' `real_size` - that is stored on the MR diff object
iself.

Fix https://gitlab.com/gitlab-org/gitlab-ce/issues/31276

Closes #29490

We see in #27387 that a call to `polymorphic_path` will cause duplicate
SELECT route calls for each merge request in a milestone. This happens
because calling `project.namespace.becomes(Namespace)` will instantiate
a new instance of a Namespace for each merge request, which causes a N+1
query on the routes table. This change caches the state of the route by
the specific class and ID, which dramatically eliminates duplicate work.

Closes #31102

Make it gracefully handle unexisting pipelines and refactor iterating
all the merge request paths.

And add some specs.

Turn ExpirePipelineCacheService into Worker so it can fetch all the
merge requests for which the pipeline runs or did run against.

- While deleting a user, some of the user's associated records are moved to the
  ghost user so they aren't deleted. The user is blocked before these records
  are moved, to prevent the user from creating new records while the migration
  is happening, and so preventing a data race.

- Previously, if the migration failed, the user would _remain_ blocked, which is
  not the expected behavior. On the other hand, we can't just stick the block +
  migration into a transaction, because we want the block to be committed before
  the migration starts (for the data race reason mentioned above).

- One solution (implemented in this commit) is to block the user in a parent
  transaction, migrate the associated records in a nested sub-transaction, and
  then unblock the user in the parent transaction if the sub-transaction fails.

If we rely on the helper functions in `GitlabRoutingHelper` for `merge_request_path`,
we end up calling a database query to look up the Namespace association for
every merge request since `entity.project.namespace` is called. By reusing the project
defined in the controller, we avoid that problem.

Partial fix to #27387

Fix
https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/10602#note_27871752

Signed-off-by: NDmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: NDavid Turner <novalis@novalis.org>

Add CHANGELOG entry

Closes #31101