Write structured metadata after Manubot AI Editor run

[falquaddoomi]

As of #71, manubot-ai-editor writes output to stdout that is scraped by the ai-revision workflow in the upcoming PR manubot/rootstock#522 to provide an authoritative response to the user about what specific model provider and model was used to revise their manuscript.

This was done for expediency, but as @d33bs mentions in #91 (review) we'd like to get away from relying on print statements and instead use Python's logging features, as raised in #66.

Since we know we'll always need to know the provider/model in the workflow, we don't want to rely on having to run the tool at a specific logging level to scrape the output. Instead, we could write a structured output, say in JSON format, that contains information about the revision run, including the provider and model that were actually used. This file could also be useful for provenance, since it'll be included in the resulting PR and thus version-controlled.

We could start with something as simple as the following:

{
  "provider": "openai",
  "model": "gpt-4o",
  "run_date": "2025-05-14 11:30:00 MDT"
}

Perhaps placing it in output, the location for other Manubot build artifacts, could work. Maybe output/ai-editor-run.json?

[falquaddoomi]

It seems that writing metadata to output will require some minor modifications to how Manubot AI Editor is invoked during the Manubot revision process. Specifically, the entrypoint class for the library ManuscriptEditor currently only gets the content and configuration folders as arguments, not the output folder.

To implement the fix I proposed, we'd need to:

• modify ManuscriptEditor to also receive the output folder and modify rootstock to pass it when invoking the editor
• Add a method to ManuscriptRevisionModel called, say, revision_metadata() that returns a dict of metadata (in GPT3CompletionModel's case, the provider and model)
• For the revision model passed into ManuscriptEditor.revise_manuscript(), call revision_metadata() and write the result to output/ai-editor-run.json