feat(ocr): add local PP-OCRv6 application capability

[ZhiXiao-Lin]

Summary

• replace the first-use OCR provider skeleton with one local PP-OCRv6_small ONNX pipeline
• add bounded model download, checksum verification, atomic installation, repair, status, and managed uninstall flows
• implement image preprocessing, detection postprocessing, recognition decoding, and typed CLI/MCP results without an off-device fallback
• include pinned model assets in release packaging checks and document licensing, provenance, storage, and runtime behavior

Dependency

This PR is intentionally stacked on #2 (recovery/builtin-ocr-first-use) at commit 2b33d9f6f043c08d79d03ceabdd2a03af6fdfb28. It upgrades that PRs built-in OCR first-use surface to the recovered PP-OCRv6 implementation. Merge #2 first; do not squash this branch onto main before its base lands.

This is a focused recovery of the PP-OCRv6 slice from interrupted development. It excludes Science, Office, signed-registry, and unrelated snapshot work.

Validation

• cargo fmt --all -- --check
• cargo test --workspace --all-features --locked
• cargo clippy --workspace --all-targets --all-features --locked -- -D warnings
• cargo test -p a3s-use-ocr --all-targets (8 unit tests and 1 contract test passed)
• cargo test -p a3s-use --all-targets
• git diff --check