Add per-question feedback fields for investigation open questions

[Neokil]

Problem

The investigation phase surfaces open questions in the artifact output, but the UI only provides a single general feedback textarea. Users have no structured way to answer individual questions — they must manually reference each question in free-form text. This makes the feedback loop awkward and means the LLM receives less structured context when rerunning the state.

Solution

1. Parse open questions out of the investigation artifact (e.g. from a defined markdown section).
2. Render a labeled text input for each question in the feedback panel.
3. Move the existing general feedback field below the per-question inputs.
4. On "Provide Feedback" submission, combine the per-question answers and general feedback into a structured message and send it to the LLM as context for the rerun.