Set up a documentation site using the Zensical framework with automatic
content extraction from README.md using markdown-code-runner.

- Add SECTION markers to README.md for content extraction
- Create docs_gen.py module to extract and transform README sections
- Add docs/ directory with pages (index, getting-started, examples, etc.)
- Add run_markdown_code_runner.py script to update all docs files
- Add GitHub Actions workflow for docs deployment
- Add zensical.toml configuration
- Add docs dependencies to pyproject.toml
- Add site/ to .gitignore for build output

Track fence delimiter character and length to correctly handle
nested fences (e.g., ```` containing ```). Previously, a simple
toggle would incorrectly treat inner ``` as a fence boundary.

Instead of escaping markers with zero-width spaces, docs templates
now contain placeholders in OUTPUT sections. The content is generated
during CI build and never committed.

Benefits:
- No escaping logic needed (removed ~70 lines of complex fence detection)
- Templates in repo are clean and copyable
- No zero-width space characters anywhere
- Simpler and more maintainable

Changes:
- docs_gen.py: Remove escaping, add --reset and --verify commands
- docs/*.md: Reset OUTPUT sections to placeholders
- docs.yml: Add verify and generate steps before build
- pre-commit: Add hook to prevent committing processed content

- Remove pip caching section (uv handles caching automatically)
- Use Python API instead of subprocess for processing files
- Use PEP 723 inline script dependencies with uv shebang

The OUTPUT placeholder now explains why we don't commit generated
content: to keep docs copyable and avoid recursion issues.

Check for <!-- PLACEHOLDER --> marker instead of full text,
so the explanatory message can be updated without breaking verification.