Add note qc feature

[jspada200]

Summary

This is a heavy but cool one! This change adds a note QC workflow. It allows users to add checks to notes with natural language then these checks run on publish. For example, you can verify a note vs the transcript.

Checks can be added via the settings UI

These checks run in the publish UI. They can update the note itself or the other fields on the note.

The checks have access to tools to browse the prodtracking software. An example of where this can be used is to find attachments to plug into the note links, tos, or CCs,

Testing

• I have tested these changes locally
• [ X ] I have run all relevant automated tests
• [ X ] I have verified this does not break existing workflows
• For changes that can be tested in UI, I have included screenshots or gif animations of the changes.

How I Tested

• Created a transcript missing info and the QC catches it
• Created a note mentioning a person who was not tagged
• Checked content of a note to ensure it contained the requested information.

[camerontarget14]

LVGTM!!! Very neat idea James!

[camerontarget14]

A few things maybe worth looking at before merge (from my AI and brushed up by me):

• generate_with_tools and generate_structured_with_tools duplicate the tool-call loop almost verbatim — consider extracting a shared helper.

• json.loads(tc.function.arguments or "{}") will throw on malformed JSON; consider wrapping and returning an error result to the tool channel so the model can recover.

• If max_iterations is hit mid tool-calling in generate_structured_with_tools, you proceed to instructor extraction without a clean final assistant turn. Worth at least a log warning, ideally a forced tool_choice="none" final pass.

As an aside:

• max_iterations bounds turns but not size — if tool_executor returns large prodtracking blobs, messages grows every iteration and you can chew through context fast. Probably worth truncating tool results before appending, or tracking approximate token usage.

[camerontarget14]

Nice!