[Feature Request] Add `format: 'html'` option to support HTML input

[vero-code]

Summary
While testing the Proofreader API (Origin Trial), I found that the model aggressively removes HTML tags from the input, treating them as noise or errors to be cleaned.

Steps to Reproduce

1. Open the official Proofreader API Playground/Demo (or run locally).
2. Input the following text containing an HTML tag and a typo:
   <div>The div is not closd
3. Run the proofread function.

Actual Result
The model returns: "The div is not closed."
The <div> tag is completely removed from the output string.

Expected Result
The model should correct the typo but preserve the HTML structure:
<div>The div is not closed.

Impact
This behavior makes the API unsuitable for use cases involving technical documentation, code editors, or rich-text environments where markup must be preserved.

[tomayac]

What you're actually asking for is for a new input type, HTML. I had filed similar feedback in webmachinelearning/translation-api#60 for the Translator API, where HTML happens to be supported surprisingly well, but more as a lucky coincidence and not because of official support.

[vero-code]

Thanks for referencing this in issue #60 — that helps clarify the broader discussion.

Having a standardized format: 'html' option would definitely make sense. For workflows involving rich-text editors or CMSs, preserving markup isn’t optional for these use cases.

[tomayac]

Could I ask you to update the Issue title to reflect the format: 'html' feature request? Thank you!

[vero-code]

Done! I've updated the title as requested.