docs: document the score scales (FineScore, Rating, Rating10, Rating20)

[YoanSallami]

Summary

Follow-up to #98: user-facing docs for the new score scales and the score_type argument, plus a fix for the broken source-snippet syntax across the docs.

• New API page Data Models API → Score Data Models rendering synalinks.src.backend.pydantic.metrics: Score, FineScore, Rating, Rating10, Rating20, plus get_score_type / normalize_score. Listed in the Data Models API index and the mkdocs nav.
• Data Models guide (guide 2): Rating row in the types table; a paragraph after the Score section introducing the sibling scales and score_type; summary bullet updated.
• Modules guide (guide 4): SelfCritique section explains the default Score reward and how score_type changes the scale (instructions + schema follow, value normalized to 0..1).
• Rewards guide (guide 13): LMAsJudge section gets a score_type=synalinks.Rating example and the normalization note.
• Source-snippet line anchors of the three edited guides recomputed; their Colab notebooks regenerated (data_models, modules, rewards). Other notebooks were left untouched even though the generator would reformat them (cell ids / unicode escaping drift vs. what's committed), to keep this PR to the docs change.
• Snippet syntax fix (57 pages): the "Source" blocks used --8 < --"path:N", which pymdownx snippets does not recognize, so the literal marker rendered instead of the code. Restored --8<-- "path:N" (and the :source section form) everywhere, and recomputed the three anchors that had gone stale (RLM agent, Deep Agent, Python Synthesis guides).

Test plan

• uv run zensical build --clean succeeds (with check_paths: true, so every snippet path resolves); the new page renders all scales/helpers and the three guide pages show the new text
• Rendered guide/example pages now contain the expanded source code and no raw --8<-- markers
• Guides still parse and pass ruff
• No em dashes introduced

🤖 Generated with Claude Code