v0.8.3

What's changed since 0.8.2

This release is driven by an oracle-verified GPU strength study on fresh Google Gemini images.

• fix(invisible): default strength raised 0.10 → 0.30. The current Google SynthID survives 0.10/0.15/0.20 and is removed only at ~0.30 (verified via the Gemini app "Verify with SynthID" on fresh Gemini images, and confirmed on a Gemini image at 0.30 with text protection on). The old 0.10 default no longer clears it — Google has hardened SynthID (the threshold has climbed 0.05 → 0.10 → ~0.30). Override per image with --strength.
• Text protection is recommended ON (the default) and does not block SynthID removal. SynthID is a global watermark, so 0.30 clears it whether or not text is protected; protection additionally keeps headings and body text readable (only the finest print still softens). --no-protect-text only trades text quality for speed.
• --no-protect-faces added on invisible/all to skip the YOLO face detector when the image has no people.
• --pipeline ctrlregen flagged EXPERIMENTAL. In testing on real images its clean-noise regeneration destroys content (hallucinated micro-text in smooth regions) with no usable middle setting, so the SDXL pipeline at ~0.30 is the shippable path.
• refactor(cli): plain-text console output (dropped the rich dependency); analysis scripts de-riched.
• Note: OpenAI / ChatGPT images carry C2PA metadata, not Google SynthID — 0.10 is plenty there. The Gemini app "Verify with SynthID" is the only valid SynthID oracle; openai.com/verify is scoped to OpenAI provenance.