Using a mountpoint as data directory causes issues when restoring, so
we need to mention this in the documentation.

See https://github.com/fog/fog-aws/issues/428#issuecomment-360869183

Closes gitlab-org/gitlab-ee#4419

https://gitlab.com/gitlab-org/gitlab-ce/commit/7f8ef19c411eceec207fef5d8459fbeaa40bf464 most likely broke the old configuration format where empty string aws_access_key_id & aws_secret_access_key were still OK. From 9.5 onwards moving backups to S3 using IAM profiles will fail if they are configured.

The new description will make it more clear that
there are 3 separate things being affected and not
just 2FA.

[skip ci]

[skip ci]

# This is the 1st commit message:

Add `copy` backup strategy to combat file changed errors

The backup Rake task used to stream data directly from the live
data directory into the backup. Under many circumstances this worked
OK. However, really active instances would experience a 'file changed
as we read it' error - especially with data like the registry. This
now copies the data first, then compresses it. It will take a bit
more disk space while the backup is in progress, but it's a
necessary thing.

# The commit message #2 will be skipped:

#	Add env var

[ci skip]

[ci skip]

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

- Location in which to add `gitlab_rails` configurations
- Reminder to run `gitlab-ctl reconfigure` after changing gitlab_rails

The amount of gems required is quite high compared to the usefulness
of the features.

Related to !4928, !6713

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

[ci skip]