Rely on Gitlab::Shell public API, not reading internal files it may use

EE brother of gitlab-org/gitlab-ce!7557 so that specs can pass on EE until the CE->EE merge (no need to pick it into stable since the CE MR will be picked already).

See merge request !887

[CE] Remove race condition while deleting groups

- EE version of gitlab-org/gitlab-ce!7528
- Related to gitlab-org/gitlab-ce#23223

See merge request !886

Added Backfill service for Geo

Geo nodes now can be synced without rsync. The sync will be triggered using our own transport method instead
and it will be triggered from the primary node at the moment a secondary node is created.

![](https://gitlab.com/gitlab-org/gitlab-ee/uploads/5a0eae3c9bf6620121e171bc07ca0a61/GeoNode_backfill.png)

Fixes #1190

cc @brodock @regisF

See merge request !861

It ensures we rely on `Gitlab::Shell`'s public API and not on its
internal. Since `Gitlab::Shell` is caching the value of its token, the
only way to get the correct token is to call
`Gitlab::Shell.secret_token`, not to read the
`Gitlab.config.gitlab_shell.secret_file` file!
Signed-off-by: Rémy Coutable <remy@rymai.me>

- Use multiple threads / database connections to:

  1. Escape the transaction the spec seems to be running
     in (`config.use_transactional_fixtures` is off, but
     `ActiveRecord::Base.connection.open_transactions` is not empty
     at the beginning of the spec.

  2. Simulate a Sidekiq worker performing the hard delete outside of the
     soft-delete transaction.

- The spec is a little clunky, but it was the smallest thing I could get
  working - and even this took a couple of hours. Let me know if you
  have any suggestions to improve it!

The intended flow is:

  Soft-delete group (sync) -> Delete group projects (async) -> Hard-delete group (async)

The soft-delete was run in a transaction, which was committed only after
the async job (for hard-deletion) was kicked off. There was a race
condition here - the soft-delete transaction could complete _after_ the
hard delete completed, leaving a soft-deleted record in the database.

This commit removes this race condition. There is no need to run the
soft-delete in a transaction. The soft-delete completes before the async
job is kicked off.

The database should not have any soft-deleted groups. Due to a race
condition (soft-delete completes after the hard-delete), soft-deleted
groups were (incorrectly) left in the database, causing issues while
trying to create a new group with the same name.

CE upstream

See merge request !880

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

Add Redis Sentinel HA docs (port from CE)

See merge request !884

[ci skip]

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

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

[ci skip]

[ci skip]

[ci skip]

[ci skip]

https://gitlab.com/gitlab-org/gitlab-ce/commit/1c994dbc05c147714479288126742f3fee158fd8#diff-4

Don't schedule backfill on create. It needs to be manually triggered now, and RepoUpdateWorker gets its own queue.

Geo nodes now can be synced without rsync. The sync will be triggered using our own transport method instead
and it will be triggered from the primary node at the moment a secondary node is created.

Fix some failing specs due to missing access_requestable trait

See merge request !7510

Fix POST /internal/allowed to cope with gitlab-shell v4.0.0 project paths

Closes #24496

See merge request !7480

fix shibboleth misconfigurations resulting in authentication bypass

This merge request fixes #22267 where a misconfigured Shibboleth `HTTP_UID` or `HTTP_EPPN` could result in users being logged into an account that did not belong to them.

See merge request !7428

Fix issue where "Without projects" filter admin area shows 0 users incorrectly. Before this fix, if any outstanding group or project invitations exist the count show as 0.

Fixes #3367

See merge request !6611

Allow commit note to be visible if repo is visible

## What does this MR do?

It enforces the `:download_code` permission in `Event#visible_to_user?` for commit notes.

Closes #23824

See merge request !7504

Stopped multiple requests with dropdowns

## What does this MR do?

Fixes an issue where the user dropdown would send 2 requests when the user opens the dropdown.

## What are the relevant issue numbers?

Closes #24131

See merge request !7505

Fixed issue boards counter border when unauthorized

## What does this MR do?

When the user is unauthorized & view issue boards, the border on the counter will not be the whole way around. This fixes that. 

## Screenshots (if relevant)

![Screen_Shot_2016-11-16_at_13.03.36](/uploads/54d97f7f4114044aad3bdc78961a31c8/Screen_Shot_2016-11-16_at_13.03.36.png)

## What are the relevant issue numbers?

Closes #23664

See merge request !7501