Expand normalization.md to the book's three-lens treatment

[dimitri-yatsenko]

explanation/normalization.md is workflow-only (~260 lines). The book's normalization chapter (30-design/055-normalization.ipynb) is much richer; several core ideas are missing from the docs:

• The three normalization lenses (Codd mathematical / Chen entity / DataJoint workflow) and their convergence on the same schema.
• A classical 1NF / 2NF / 3NF baseline + functional dependencies — without it, the repeated "workflow normalization is stricter than 3NF" claim has no referent.
• The Entity Normalization Principle — attributes describe the entity directly, completely, non-optionally — including relationship-sets-as-entities. (Docs currently substitute a narrower "Intrinsic Attributes Principle.")
• The e-commerce Order example — a table that passes 3NF and entity normalization yet fails workflow normalization (NULLs + UPDATEs). The clearest, domain-neutral motivator; docs use only mouse/cage.
• The UPDATE-as-design-smell heuristic ("can this be pure INSERT + DELETE?") in the explanation tier (currently only in a reference spec).
• The tidy-data mapping (Wickham → 1NF + entity normalization).

Proposal: expand normalization.md into the three-lens structure; keep the existing (strong) mouse/cage example as the DataJoint worked example.

From an internal datajoint-book → docs coverage review (priority P3).