[NisargOza]

CCO listing runs previously emitted empty placeholder records for failed pages, counted them toward the document limit, and used source-specific CLI dispatch that could stop --source all early. This ports the scraper from #17 onto the shared registry and listing contract on current main, so failed pages are logged/skipped and only successful documents count.

• Register CCO alongside NICE and MedlinePlus; reject HTTP error pages and unreadable HTML, canonicalize numeric detail URLs, and retain the master listing's Load more pagination.
• Declare Playwright in the optional cco extra, gate imports, and document Chromium installation. Normal runtime imports work without Playwright.
• Preserve version/status/authors/PDF links and mark the extracted HTML as summary; linked PDF bodies are not converted.
• Add local-output/environment ignore rules and document branch names without the codex/ prefix in AGENTS.md. Keep the existing uv lockfile tracked.

Continues #17 and addresses its review comments: dependency declaration, all-source dispatch, and failed documents counted as success. The original PR remains open for its review history.

Validation: uv run ruff check ., uv run ruff format --check ., uv run pytest -q (70 passed), and uv build --package amfv-datasets. A fresh install without Playwright successfully imports the CLI and reports actionable installation instructions when CCO is invoked. Live checks scraped one explicit URL and two listing documents; the complete corpus was not run.