allow overriding `{host,bootstrap}-containers` entrypoint command

[kasimeka]

related issue: bottlerocket-os/bottlerocket#1250

What I'd like:
a way to create a host-container with a custom entrypoint command, so i can -for example- run any container from any source with a sleep infinity entrypoint that lets it stay alive and allows me to connect to it with apiclient exec without having to create a new OCI image to hardcode a new entrypoint in.

[kasimeka]

i made draft PRs (core-kit, settings-sdk) for the change since it's simple enough and i think they might be clearer than discussing the approach in plain english. the change adds a command array to host-containers and bootstrap-containers settings that replaces the container's ProcessArgs (its entrypoint command) when it's set. the host-ctr and systemd unit changes are working as expected but i just need a tiny bit of guidance on how to get the new field accepted by the apiserver's json deserializer because it seems that something's missing and im getting the following error:

$ apiclient set -j '{ "settings": { "host-containers": { "mycontainer": {
  "enabled": true,
  "source":"328549459982.dkr.ecr.us-east-1.amazonaws.com/bottlerocket-bootstrap:v0.2.4",
  "command": ["sleep","infinity"],
  "user-data": "ZWNobyBoZW5sb28K" } } } }'
Failed to change settings: Failed PATCH request to '/settings?tx=apiclient-set-i7yE9hlwIQRjhVxr': Status 400 when PATCHing /settings?tx=apiclient-set-i7yE9hlwIQRjhVxr: Json deserialize error: unknown field `command`, expected one of `source`, `enabled`, `superpowered`, `user-data` at line 1 column 44

[bcressey]

I've been somewhat down on this kind of feature in the past, because I don't want to see the host container interface grow in complexity to rival what container orchestrators support.

However, your use case makes a lot of sense to me. I've also said that if we'd had apiclient exec from the start, the admin container in particular would be very different - it'd just be an unmodified Amazon Linux image that ran "sleep 1000" and waited for users to exec into it, and we'd have separate host containers for ssh and serial console access.

With that in mind, this feels like a way to make apiclient exec more useful and Bottlerocket less opinionated about the distro you use for your debugging tools of choice, which is great.