* Adds support for packages to the backup and restore rake tasks

Changelog: added

Reuses standard file upload storage to cover Terraform
states backup

Adds a seed for generating terraform state files
under each project

Manual testing (in local GDK env):

- Verfied seed addition via `bin/rake db:seed_fu FILTER=terraform`
- Verified seed addition is repeatable, with no-ops if versions
  pre-exist
- Created backup: `bin/rake gitlab:backup:create`
- Verified output shows up:

```
2021-06-05 23:48:54 +0530 -- Dumping terraform states ...
2021-06-05 23:48:54 +0530 -- done
```

- Verified tar contains terraform_state.tar.gz
- Verified terraform_state.tar.gz carries content
- Destroyed all records: `Terraform::State.destroy_all`
- Deleted all files: `rm -rf shared/terraform_state/*`
- Performed restore: `bin/rake gitlab:backup:restore`
- Verified `Terraform::State.all` shows back older records
- Verified association of builds to some of these records
- Verified `shared/terraform_state/` is repopulated
- Verified `Terraform::StateVersion.each do |sv| pp sv.file.read end`
  works and shows the seed-added data
- Verified skipping of `terraform_state` keyword works:

```
2021-06-06 00:11:33 +0530 -- Dumping terraform states ...
2021-06-06 00:11:33 +0530 -- [SKIPPED]
```

Verified with same workflow as above (backup, destroy, delete,
then restore) that skipped backup variant does not contain
`terraform_state.tar.gz` and does not show up those files after
restore

Miscellany:

- Adds a missing test for LFS backups
- Adds File operation fixes to neighboring tests
  so when they fail RSpec is able to show diffs
- Modified test validates unknown SKIP values not
  breaking backups
  - This is useful to catch any regressions during
    upgrades where pre-installs scripts may refer
    to a new SKIP value that isn't recognized by
    the old version

Changelog: added

The old rules meant that any future changes to these directories would
require the use of `git add -f` to be staged. This is problematic for
the upcoming FOSS-only mirror of GitLab. Any changes to be synced to
this mirror are staged using `git add --all .`, which due to the Git
ignore rules would not add the tmp/ and shared/ directories. This then
leads to various specs failing because they expect certain
sub-directories of these directories to be available.

This commit changes the rules so that we include the .gitkeep files,
ignore other files, and completely ignore anything in tmp/test. This
requires that the rules are specified in tmp/.gitignore and
shared/.gitignore, instead of the top-level .gitignore, without this I
could not get Git to include the .gitkeep files properly.

- The pages are created when build artifacts for `pages` job are uploaded
- Pages serve the content under: http://group.pages.domain.com/project
- Pages can be used to serve the group page, special project named as host: group.pages.domain.com
- User can provide own 403 and 404 error pages by creating 403.html and 404.html in group page project
- Pages can be explicitly removed from the project by clicking Remove Pages in Project Settings
- The size of pages is limited by Application Setting: max pages size, which limits the maximum size of unpacked archive (default: 100MB)
- The public/ is extracted from artifacts and content is served as static pages
- Pages asynchronous worker use `dd` to limit the unpacked tar size
- Pages needs to be explicitly enabled and domain needs to be specified in gitlab.yml
- Pages are part of backups
- Pages notify the deployment status using Commit Status API
- Pages use a new sidekiq queue: pages
- Pages use a separate nginx config which needs to be explicitly added

This reverts commit 3e991230.

# Conflicts:
#	app/models/project.rb

- The pages are created when build artifacts for `pages` job are uploaded
- Pages serve the content under: http://group.pages.domain.com/project
- Pages can be used to serve the group page, special project named as host: group.pages.domain.com
- User can provide own 403 and 404 error pages by creating 403.html and 404.html in group page project
- Pages can be explicitly removed from the project by clicking Remove Pages in Project Settings
- The size of pages is limited by Application Setting: max pages size, which limits the maximum size of unpacked archive (default: 100MB)
- The public/ is extracted from artifacts and content is served as static pages
- Pages asynchronous worker use `dd` to limit the unpacked tar size
- Pages needs to be explicitly enabled and domain needs to be specified in gitlab.yml
- Pages are part of backups
- Pages notify the deployment status using Commit Status API
- Pages use a new sidekiq queue: pages
- Pages use a separate nginx config which needs to be explicitly added

- Offloads uploading to GitLab Workhorse
- Use /authorize request for fast uploading
- Added backup recipes for artifacts
- Support download acceleration using X-Sendfile