feat(api-server): Enable charm configuration of launcher and driver images

[orfeas-k]

Enable charm configuration of launcher and driver images used during pipeline steps. Those can be configured thourgh values launcher-image and driver-image. When unset, KFP uses the default values defined. This is based on upstream implementation introduced in kubeflow/pipelines#10269.

Closes #452

Testing

This was tested with upstream image gcr.io/ml-pipeline/api-server:2.0.5. The rest of kfp used the ROCKs.

It can be tested that pipeline steps are using the proper images by configuring the charm and then running a pipeline.

juju config kfp-api launcher-image="fake-image"
juju config kfp-api driver-image="fake-image"

You should see pods created by the pipeline run that spin up containers using the images provided in the config options.

Note that:

• Each configuration should be tested separately. A pod using the launcher-image is only spun up after (at least some) driver-image pods are completed. Thus, launcher-image configuration can only be tested when driver-image config is healthy.
• When rerunning the same exact pipeline, KFP will use the launcher spun up previously, if one was successfully created, instead of spinning a new one (thus the previously used launcher-image).

[ca-scribner]

changes lgtm but we need to update the tools/get-images.sh file as commented

[NohaIhab]

I synced with @orfeas-k to test this, and it was working correctly for me!
We agreed to add to the PR the instructions that we tested with.

update: instructions now in the PR description

[kimwnasptd]

Deployed as well the Charm and managed to change the images successfully.

already addressed and approved by someone else