v0.11.3

Changes since v0.11.2

• C2PA reading migrated to the official c2pa-python library (now a core dependency, MIT/Apache). extract_c2pa_info uses the spec-tracking Reader first (any container, whole manifest store including ingredient manifests) and falls back to the hand-rolled caBX parser for blobs the validator rejects. Non-PNG (JPEG/WebP/AVIF) manifests are now parsed structurally. Validated on a 25,725-image corpus: 0 regressions, 384 manifests newly parsed, 3 false Adobe/Microsoft->Google attributions fixed. Dead from c2pa import has_c2pa_metadata import removed.
• lower(strength): vendor-adaptive floor dropped to OpenAI 0.10 / Google 0.15.
• fix(visible): inpaint a mid-tone Gemini sparkle instead of leaving a dark diamond.
• fix(invisible): preserve native output dimensions.
• perf(identify): decode the image once for all visible-mark detectors.
• fix(cli): preserve SynthID uncertainty in the no-visible-mark message.
• chore(deps): bump the minor-and-patch group (5 updates).