This patch implements the full zero downtime upgrade path recommended by
upstream into the helm chart, which should provide a smooth experience
for upgrades.

Currently the jobs still had the old S3 dependant affinity magic going,
which seems violates the rule of least surprise (in my perspective),
therefore removing the magic affinity adjustments and instead go with
using the same affinities defined for all web Pods.

This patch replaces removes various variables from the configmap and
moves them to Kubernetes secrets to keep things organised.

Cherry-Pick upstream: https://github.com/mastodon/chart/pull/35

Cherry-Pick of https://patch-diff.githubusercontent.com/raw/mastodon/chart/pull/5.patch



[PATCH] Add the PREPARED_STATEMENTS environment variable and set it
 to its default value Being able to control it can be useful in situations
 where prepared statements don't work as expected. Once such case can be
 pgbouncer as provided by DigitalOcean for connection pooling.

Co-authored-by: Cees-Jan Kiewiet <ceesjank@gmail.com>

This patch resolves a potential issue with the streaming endpoint, which
is different in the docs, than in the helm chart. This is integrating the
upstream issue into this chart, which should resolve potential issues
with non-default frontends: https://github.com/mastodon/mastodon/pull/19896

Currently the rpm-ostree upgrade command requires some form of
credentials or refuses to download the image. rpm-ostree rebase works
fine.

Hope this will be fixed so, for now a read-only robot account for quay…

This patch enables digest pinning for renovate updates, this should help
to keep everything updates even while using mutable tags, such as
`stable` for CoreOS.

This patch provides the first release of kOS, a new Fedora CoreOS remix,
that is focused on the current setup of the SI infrastructure.

It basically implements all the little hacks and adjustments that are
currently active on the (existing) Fedora Server nodes running the
infrastructure and will provide a better replacement as soon as they are
put in production.

This patch should group up the rook operator and cluster updates into one
Merge Requests instead of multiple, in order to reduce potential
incompatiblities.

For a reason I haven't fully figured out yet, the kubectl plugin
discovery in kaniko doesn't work like in podman or docker. This is
probably due to how they invoke shells for the container build. Either
way, this adjustment should hopefully fix it.

This patch explicitly configures the storage devices to use and adjusts
the replication factor for some volumes to be a bit easier on the overall
cluster.