Between the time when I ported the Action's code for parsing config files to the CLI and when we actually switched the Action to using the new code, the Action's code gained an interesting new effect that was not in the ported code: If the feature flag for ATM was set the Action would, while it was parsing the file, also add the ATM packs to it if they weren't already present. When we switched over to using the parsing in the CLI, this effect was lost and we stopped running ATM queries (!!).

This PR is a rather hacky fix that addresses this by recording (when the Action parses the config file - which it still does even if it later doesn't need it) whether it injected the ML queries. Then, when we pass the full config to the CLI we also augment it with the ML queries if we injected them before. It's not how I would like to do it, but there's a couple of reasons we can't do something more elegant:

1. We don't want to wait for a CLI release to get things working again, so moving this injection into the CLI isn't an option. Additionally, the CLI would need to know whether the feature flag is set, and also the version of the ML pack to inject (which during the beta we want to record in the Action due to its shorter release cycle), so we'd have to pass a couple more flags in.
2. We can't stop injecting into the Action's parsed version of the config file as well because old CLIs still use this. Plus, the telemetry looks here to know whether the ML queries are enabled.

This code is also missing tests, we have an internal issue tracking this. I'll manually test this for now, and before we touch this code again we should have an integration test that actually runs ATM queries, so we don't accidentally disable them again.

Merge / deployment checklist

• Confirm this change is backwards compatible with existing workflows.
• Confirm the readme has been updated if necessary.
• Confirm the changelog has been updated if necessary.