feat(efcore): Weasel model → EF MigrationOperation translation layer - #375

Closes #365. First implementation phase of the EF Core migration generation epic (#371), building directly on the verified #364 spike results (attribute-only migrations with empty target models are first-class in the EF toolchain).

What's here

New MigrationOperationTranslation in Weasel.EntityFrameworkCore — the reverse direction of MapToTable: walk Weasel's schema model and produce EF Core MigrationOperation instances, the intermediate representation the C# migration file emitter (#366) will render as MigrationBuilder calls.

• Input is the provider-neutral surface (ITable, ITableColumn, ITableIndex, ForeignKeyBase, SequenceBase) — the project keeps zero concrete-provider references. Two small additive changes to Weasel.Core expose what was missing: ITable.Columns and ITableIndex.Columns (implicitly satisfied by every provider's existing concrete members).
• Raw store types everywhere: every column carries ColumnType so EF's CLR type mapping is bypassed and the resulting DDL matches Weasel's own exactly. The ClrType is a best-effort inverse map used only for the table.Column<T>(...) generic argument in emitted C#.
• Structure: CreateTable with nested columns / primary key / check constraints / foreign keys, one CreateIndex per index, EnsureSchema per non-default schema (deduplicated; public/dbo become null Schema, matching EF's own scaffolding), CreateSequence from SequenceBase (StartWith/IncrementBy for HiLo).
• Provider mapping (per the issue's table): identity → Npgsql:ValueGenerationStrategy / SqlServer:Identity annotations; computed columns → ComputedColumnSql + IsStored (always stored on PostgreSQL); index filter → first-class Filter, includes/method → provider annotations; CascadeAction → ReferentialAction with SQL Server Restrict ≡ NoAction, mirroring the mapDeleteBehavior normalization from the EF→Weasel direction.
• Raw-SQL fallback: everything EF cannot model routes through SqlOperation carrying the object's own WriteCreateStatement output — non-table/non-sequence schema objects (functions, sprocs, table types) automatically, plus anything matched by the ForceRawSql hook (partitioned tables being the canonical case; callers with provider references downcast there, same pattern as the harness's customizeTables). Expression indexes throw NotSupportedException with guidance to use the hook.
• ToDropMigrationOperations builds Down() bodies: reverse-order DropTable/DropSequence/raw drops. Schemas are deliberately never dropped (they may be shared with Marten/Wolverine).
• Npgsql identity annotation value is the string "IdentityByDefaultColumn" in-memory (this project can't reference the provider enum); the EF migration generation: C# migration file emitter + stub DbContext #366 emitter renders it as the proper NpgsqlValueGenerationStrategy literal, and callers feeding operations directly to the Npgsql SQL generator can overwrite NpgsqlIdentityAnnotationValue with the real enum.

Renames (RenameColumnOperation/RenameTableOperation) belong to the incremental TableDelta phase (#367) per the issue's design note.

Testing

13 new DB-free unit tests covering: full PG table shape (raw types, nullability, defaults, identity annotation, PK, checks), SQL Server identity, computed columns both providers, index filter/includes/method, expression-index rejection, FK referential actions incl. the Restrict normalization split, default-schema omission, sequence translation, ForceRawSql partitioned-table fallback, missing-Migrator error, drop ordering, and PreserveIdentifierCase flow-through.

End-to-end validation (operations → SQL → schema comparison) deliberately belongs to #369. Full local suites green: Core 21, SQLite 361, PostgreSQL 767, SQL Server 302, EF Core (PG+SS) 75.

🤖 Generated with Claude Code