Make methods return ActiveRecord Relations instead of Arrays.

This ensures that if a project is later re-created using the old path it
doesn't end up re-using the old cache. This also ensures we don't keep
the cache around until its expired by Redis itself.

Fixes gitlab-org/gitlab-ce#13790

This keeps all the cache expiration code in a single file/class instead
of spreading it all across the codebase.

Fixes #13367.

The "default_branch" argument is never used and the "project" argument
isn't optional.

- Add `avatar_url`, `description`, `git_ssh_url`, `git_http_url`,
  `path_with_namespace` and `default_branch` in `project` in push, issue,
  merge-request and note webhooks data
- Deprecate the `ssh_url` in favor of `git_ssh_url` and `http_url` in
  favor of `git_http_url` in `project` for push, issue, merge-request and
  note webhooks data
- Deprecate the `repository` key in push, issue, merge-request and
  note webhooks data, use `project` instead

This caches the output of the following methods:

* Repository#empty?
* Repository#has_visible_content?
* Repository#root_ref

The cache for Repository#has_visible_content? is flushed whenever a
commit is pushed to a new branch or an existing branch is removed.
The cache for Repository#root_ref is only flushed whenever a user
changes the default branch of a project. The cache for Repository#empty?
is never explicitly flushed as there's no need for it.

By simply loading the first event from the already sorted set we save
ourselves extra (slow) queries just to get the latest update timestamp.
This removes the need for Event.latest_update_time and significantly
reduces the time needed to build an Atom feed.

Fixes gitlab-org/gitlab-ce#12415

By instead using a sub-query we save ourselves the overhead of loading
any data into memory only to pass it on to another query.

This greatly improves the performance of Project#external_issue_tracker
by moving most of the fields queried in Ruby to the database and letting
the database handle all logic. Prior to this change the process of
finding an external issue tracker was along the lines of the following:

1. Load all project services into memory.
2. Reduce the list to only services where "issue_tracker?" returns true
3. Reduce the list from step 2 to service where "default?" returns false
4. Find the first service where "activated?" returns true

This has to two big problems:

1. Loading all services into memory only to reduce the list down to a
   single item later on is a waste of memory (and slow timing wise).
2. Calling Array#select followed by Array#reject followed by Array#find
   allocates extra objects when this really isn't needed.

To work around this the following service fields have been moved to the
database (instead of being hardcoded):

* category
* default

This in turn means we can get the external issue tracker using the
following query:

    SELECT *
    FROM services
    WHERE active IS TRUE
    AND default IS FALSE
    AND category = 'issue_tracker'
    AND project_id = XXX
    LIMIT 1

This coupled with memoizing the result (just as before this commit)
greatly reduces the time it takes for Project#external_issue_tracker to
complete. The exact reduction depends on one's environment, but locally
the execution time is reduced from roughly 230 ms to only 2 ms (= a
reduction of almost 180x).

Fixes gitlab-org/gitlab-ce#10771

Properly ensure that the token exists and add defensively check for a
non-nil value.

Closes #4294

The introduction of ActiveJob and `deliver_now` in 7f214cee caused a race
condition where the mailer would be invoked before the project was committed
to the database, causing the transfer e-mail notification to show the old
path instead of the new one.

Closes #4670

This is needed to support Huboard and a generally useful value.