Re-enqueue audio render jobs for articles whose body_text was rewritten (typically by the PII sanitizer scrub) after the canonical audio was already cached in R2. The audio in R2 still narrates the pre-scrub text — we need fresh renders so listeners hear the cleaned body.

Resolution per article:

• Pick the voice_id from the most-recent done RenderJob (the voice the audio listeners are currently hearing). Falls back to the publisher's preferred_voice_id, else "kira" — the platform default and a member of VALID_VOICE_IDS.
• Use the article's current cleaned body_text.
• Inherit speed from the most-recent done job; default 1.0.
• Enqueue with job_priority='low' so this backfill doesn't starve foreground listener intake.

Idempotent: an article is skipped if its most-recent done RenderJob already rendered the current body_text (i.e. text matches).