Auto Improve And Evaluate

Single-shot auto-improve + evaluate. The studio calls this when a JD first lands (Generate / Paste / Upload). Two LLM calls server-side:

1. auto_improve_jd → returns revised_jd + auto_applied_changes
2. evaluate_jd_quality(revised_jd) → returns score + remaining issues

The frontend then shows: improved JD in textarea, "We auto-applied N improvements" list, and a needs-input section for any remaining issues that require recruiter values (salary / company / location).

Returns: { overall: 0-100, # score on the REVISED JD dimensions: {...}, # remaining issues only revised_jd: "<full text>", auto_applied_changes: ["..."], original_text_length: int, # for "X chars → Y chars" UI revised_text_length: int, }