feat(transform): object-level schema routing via SchemaRouter

[pyramation]

Summary

Generalizes the schema transform from a whole-schema Map<oldSchema, newSchema> to a SchemaRouter that resolves a target schema per object occurrence — (schema, name, namespace) — instead of per schema. This makes a single source schema fan out at the object level: e.g. route a table to one schema and a function to another, with every cross-object reference (SQL bodies, PL/pgSQL bodies, DROP/revert statements, types, params, return types) rewritten to each object's routed target.

Motivation: downstream (pgpm apply modules) needs to cherry-pick objects out of a reusable package into consumer-selected schemas — e.g. a shared helper function in reporting that reads a per-tenant tenant_a.accounts table, from a single users-schema source. A Map<string,string> can only send all of users to one place.

Model

SchemaRouter mirrors the routing shape already used by qualifyUnqualified (relations / functions / types buckets → matching pg_class / pg_proc / pg_type):

new SchemaRouter({
  users: {
    schema: 'tenant_a',                       // schema-level default (optional)
    functions: { account_count: 'reporting' } // per-object override
  }
});

Resolution is object-route-first, then the schema-level default, then leave unchanged. Namespaces are independent (a table widget and a function widget can route to different schemas). SchemaRouter.fromSchemaMap(...) reproduces the historic whole-schema behaviour exactly.

Changes

• router.ts (new) — SchemaRouter, SchemaRoute, RouteSpec, ObjectNamespace, RouteNamespace.
• Threaded an object namespace through transformNameList / transformRelation; annotated create & reference sites (functions, types, relations) and DROP … (namespace derived from removeType, so revert scripts route correctly).
• LANGUAGE sql bodies (opaque String nodes) are rewritten AST-precisely (parse → walk → deparse) when object routes are present; pure whole-schema routes keep the existing quoting-preserving string pass — so existing behaviour is byte-for-byte unchanged.
• validateNoUntransformedSchemas now enforces the leftover check only for fully-moved schemas (those with a schema-level default); partially/object-only routed schemas may legitimately retain references.
• transformSql / transformSqlStatement accept Map<string,string> | SchemaRouter (SchemaMappingInput). assumeSchemasExist and SchemaTransformResult reporting are preserved.

Backwards compatibility

Map<string,string> remains fully supported — it is coerced to a schema-level-only router. All 160 existing tests pass unchanged.

Tests

12 new tests (__tests__/router.test.ts): router unit behaviour (override/fallback/namespace independence/fully-moved reporting), table+function fan-out from one schema, cross-object refs in SQL and PL/pgSQL bodies, DROP routing by namespace, assumeSchemasExist with a router (CREATE SCHEMA IF NOT EXISTS), and Map/router output parity. Round-trip validation enabled on the full-module test.

Test Suites: 5 passed, 5 total
Tests:       172 passed, 172 total

Notes

• pnpm run build succeeds. pnpm run lint fails repo-wide with a pre-existing ESLint 9 vs legacy .eslintrc.json mismatch (couldn't find an eslint.config.(js|mjs|cjs)), unrelated to this change.
• Consumed downstream by constructive-io/constructive (pgpm object-level apply routing) — PR linked separately.

Link to Devin session: https://app.devin.ai/sessions/025fb88043964fdbb335ac5e39df2478
Requested by: @pyramation

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring