Use separate containerd configs in aws-k8s and aws-dev

[tjkirch]

We were failing to write the containerd config in the dev image because
pod-infra-container-image wasn't set -- because it lives in the Kubernetes
settings we don't have in the dev image.

We don't need a pod-infra-container-image (or similar) setting in the dev image
because we use containerd through Docker which doesn't use it. So, the
clearest change is to split the containerd config file, removing the irrelevant
cri-plugin settings from the dev version, including the
pod-infra-container-image setting.

---

Testing done:

Built and launched an aws-k8s image; it connected to my cluster and pods worked. Logging in, I see the config files:

bash-5.0# ls -l /usr/share/templates/containerd-config-toml*
-rw-r--r-- 1 root root 337 Dec 16 20:23 /usr/share/templates/containerd-config-toml_aws-dev
-rw-r--r-- 1 root root 920 Dec 16 20:23 /usr/share/templates/containerd-config-toml_aws-k8s

...and confirmed that /etc/containerd/config.toml is the k8s version with sandbox_image set based on pod-infra-container-image.

Then I built an aws-dev image, connected and confirmed Docker containers still run OK. Logging in, I see the same two config file templates, and confirmed that /etc/containerd/config.toml is the new shorter version for dev:

bash-5.0# cat /etc/containerd/config.toml 
version = 2
root = "/var/lib/containerd"
state = "/run/containerd"
disabled_plugins = [
    "io.containerd.snapshotter.v1.aufs",
    "io.containerd.snapshotter.v1.zfs",
    "io.containerd.snapshotter.v1.devmapper",
]

[grpc]
address = "/run/containerd/containerd.sock"

[plugins."io.containerd.internal.v1.opt"]
path = "/opt/containerd"

[jhaynes]

I don't love multiple config files and the .spec install-splosion that this causes... but since both of those are fairly straightforward to undo, ⛵️ 🚀.

[etungsten]

🚢