Multistep bootstrap - WIP

[enxebre]

This decouples the main bootstrapping process into 5 steps:
To deploy: mkdir cluster-aws; cd cluster-aws and create your terraform.tfvars

• Generate tls content
  terraform apply -state tls.tfstate ../steps/tls/
• Generate assets e.g operator manifests (and NCG ignition config manifests, this will need to be consolidated with NCG/chain model generation)
  terraform apply -state assets.tfstate ../steps/assets/
• Generate Ignition config for bootstrap node
  terraform apply -state ignition.tfstate ../steps/ignition/
• Deploy the bootstrap node which runs the ncg
  terraform apply -state bootstrap.tfstate ../steps/bootstrap/
  wait for ncg to run, then:
  terraform destroy -force -state=bootstrap.tfstate -target aws_route53_record.tectonic_ncg ../steps/bootstrap/
• Scale the cluster to the target size by deploying the joining nodes consuming from the NCG
  terraform import -state=joining.tfstate -config=../steps/joining/ aws_autoscaling_group.masters cluster-aws-masters
terraform apply -state joining.tfstate ../steps/joining

This also removes all leader election logic.

TODO:

• To avoid skew between the bootstrap node and ncg nodes, currently the bootstrap node gets its s3 ignition config (identical to NCG config) through an http endpoint. This needs to be addressed before merging.
• To stop NCG config manifests to be deployed to other platforms (doing it for convenience at the minute)
• Each step before 4 bootstrap and 5 joining are almost cloud agnostic and should be moved from terraform to best suited specific tools
• Once Cli tool #2806 lands it will need to be added support for this multistep workflow

[coreosbot]

Can one of the admins verify this patch?

[enxebre]

cc @alexsomesan

[enxebre]

Closing this and breaking this down into separate smaller PRs starting with:
Assets generation - #2905
Two steps bootstrap #2924