Fix empty "rendered" with FCC 1.1 and snippets

[bendrucker]

When pretty_print = false is used along with FCC 1.1 configuration and snippets, the result is an empty (default) Ignition document. Adapting a case from the tests:

terraform {
  required_version = ">= 0.13"

  required_providers {
    ct = {
      source  = "poseidon/ct"
      version = "~> 0.7.0"
    }
  }
}

data "ct_config" "fedora-coreos-snippets" {
  pretty_print = false
  strict = true
  content = <<EOT
---
variant: fcos
version: 1.1.0
passwd:
  users:
    - name: core
      ssh_authorized_keys:
        - key
EOT
	snippets = [
<<EOT
---
variant: fcos
version: 1.1.0
systemd:
  units:
    - name: docker.service
      enabled: true
EOT
	]
}

output "ignition" {
  value = data.ct_config.fedora-coreos-snippets.rendered
}

The result is:

$ terraform apply -auto-approve
data.ct_config.fedora-coreos-snippets: Refreshing state... [id=475385134]

Apply complete! Resources: 0 added, 0 changed, 0 destroyed.

Outputs:

ignition = {"ignition":{"config":{"replace":{"verification":{}}},"proxy":{},"security":{"tls":{}},"timeouts":{}},"passwd":{},"storage":{},"systemd":{}}

This is a regression from 0.6, which outputs the expected rendered result:

$ terraform apply -auto-approve
data.ct_config.fedora-coreos-snippets: Refreshing state... [id=475385134]

Apply complete! Resources: 0 added, 0 changed, 0 destroyed.

Outputs:

ignition = {"ignition":{"config":{"replace":{"verification":{}}},"proxy":{},"security":{"tls":{}},"timeouts":{},"version":"3.1.0"},"passwd":{"users":[{"name":"core","sshAuthorizedKeys":["key"]}]},"storage":{},"systemd":{"units":[{"enabled":true,"name":"docker.service"}]}}

I added test cases where pretty_print is set to false, which demonstrated that this is only an issue with FCC version 1.1. The cause turns out to be the wrong Ignition object being passed to json.MarshalIndent, specifically ign instead of ign31. Passing that in to match json.Marshal results in a passing test.

[bendrucker]

I've also pulled out a marshalJSON func that handles switching between Marshal and MarshalIndent to try to prevent similar bugs

[bendrucker]

Instead of creating entirely separate test configs, I could also convert containerLinuxSnippetsResource et al into functions that take a pretty argument and use fmt.Sprintf to render the configuration. Wanted to avoid modifying other test cases initially but happy to do that now if you'd prefer.

[dghubble]

I think its fine. Eventually all this custom merging should be handled upstream in fcct, which would avoid all the branches motivating this exhaustive testing. coreos/butane#120

[dghubble]

Thanks!