Create the database schema

[anettkva]

Write the Prisma schema for the whole data model and create the migrations. Follow docs/DATA_MODEL.md and the search design in docs/API.md ("Search without raw SQL").

Done when:

• All tables of the data model exist: apparatus (with position), element groups, variant groups, elements, vaults, coaches, gymnasts, gymnast element lists, routines and routine items
• The catalog keeps the readable ids from data/, and user data gets generated ids (cuid)
• Lists are stored as rows: routine_elements (with a position from 1, unique per routine) and gymnast_elements (no duplicates per gymnast). Each row points to an element or a vault, and a CHECK makes sure exactly one is set
• Elements and vaults have a nullable variant_group_id and removed_at. Routines have created_at and updated_at
• value and d_value are Decimal(3, 1), with CHECKs on their ranges
• User data cascades on delete, and catalog rows in use cannot be deleted
• The search tables CatalogEntry (one row per code) and CatalogWord exist, with the indexes for filtering and sorting, and the C collation on catalog_words.word so the prefix search can use its index
• prisma migrate dev reports no changes after the migrations are applied
• docs/DATA_MODEL.md describes how the model is stored in PostgreSQL, and agrees with schema.prisma on every table and field
• backend/src/schema.graphql matches the decisions from the review: the gymnast's element list (Gymnast.elements, addGymnastElement, removeGymnastElement), removed on catalog items, and the difficulty filter as a list of values with sorting by difficulty only within an apparatus