Reduce Chef configuration times by at least 25%

As discussed in the DR blueprint there are optimizations that will help us improve our recovery time objective (RTO) in the case of a disaster. Currently when we provision we provision always from scratch on Chef managed nodes.

One way to improve this would be to use images, and some of this was already done for Gitaly (https://gitlab.com/gitlab-com/runbooks/-/blob/master/docs/gitaly/gitaly-os-upgrade.md) though we are no longer running the scheduled pipeline to generate them.

There are more options than just using images, an arbitrary goal is set to improve our configuration times by 25%.

Note: The work here maps to key result https://gitlab.com/gitlab-com/gitlab-OKRs/-/work_items/5707+