The refs hash is used to determine what branches and tags have a commit
as head in the network graph. The previous implementation depended on
Rugged#references. The problem with this implementation was that it
depended on rugged, but also that it iterated over all references and
thus loading more data than needed if for example the project uses CI/CD
environments, Pipelines, or Merge Requests.

Given only refs are checked the network cares about the GraphHelper#refs
method has no need to reject those, simplifying the method.

Closes gitlab-org/gitaly#880

From https://gitlab.com/gitlab-org/gitlab-ce/issues/36760

Was reverted in
https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/14373

Callers expect a URL.

This fixes the avatars in commit lists when a user has no avatar and
Gravatar is disabled.

Closes #38715

fix bug in webpack_helper in which force_same_domain argument was ignored breaking local rspec tests

Search query is especially slow if a user searches a generic string
which matches many records, in such case search can take tens of
seconds or time out. To speed up the search query, we search only for
first 1000 records, if there is >1000 matching records we just display
"1000+" instead of precise total count supposing that with such amount
the exact count is not so important for the user.

Because for issues even limited search was not fast enough, 2-phase
approach is used for issues: first we use simpler/faster query to get
all public issues, if this exceeds the limit, we just return the limit.
If the amount of matching results is lower than limit, we re-run more
complex search query (which includes also confidential issues).
Re-running the complex query should be fast enough in such case because the
amount of matching issues is lower than limit.

Because exact total_count is now limited, this patch also switches to
to "prev/next" pagination.

Related #40540