Improve idempotency when applying changes to resources

[invidian]

Containers engine already updates current state on every applied change, but it is not used until all execution succeeds. If something fails in the middle of the process, current state is currently thrown away by both binaries and Terraform provider, which then requires manual cleanup before next run.

To improve that, state should be persisted regardless of execution error, so on the next run, engine can pick it up and continue from where it left off. This allows user to resolve encountered issue and continue deploying.

[invidian]

#78 migrates flexkube_etcd_cluster and flexkube_apiloadbalancer_pool Terraform resources to use new Terraform scheme and helpers. While testing that, I found out many places, which required improvements related to idempotency, and all of them are now fixed.

#65 migrated all CLI tools to use new cli package, which also improves idempotency.

Once all Terraform resources are migrated to new scheme, this issue can be closed.

[invidian]

#82 migrated kubelet and #81 migrated bootstrap controlplane, so while the process is still not ideal (e.g. kubelet flags are not ensured), it improved a lot and it no longer loses the information about created containers on failure, which was the main problem here. Closing for now.