fix(codegen): include plugin-injected fields in condition/orderBy types + VectorFilter

[pyramation]

fix(codegen): support pgvector types and plugin-injected condition/orderBy fields

Summary

The codegen only iterated over table.fields (database columns) when building condition and orderBy types, so any fields/values injected by Graphile plugins (e.g., embeddingNearby from VectorSearchPlugin, EMBEDDING_DISTANCE_ASC/DESC orderBy values) were silently dropped from generated TypeScript.

This PR fixes four things:

1. scalars.ts — Adds Vector: 'number[]' to SCALAR_TS_MAP and Vector: 'VectorFilter' to SCALAR_FILTER_MAP so pgvector's Vector scalar resolves correctly.
2. Plugin-injected condition fields — buildTableConditionProperties now accepts an optional TypeRegistry and merges extra fields present in the schema's condition type that aren't derived from table columns. A new collectConditionExtraInputTypes helper ensures referenced types (e.g., VectorNearbyInput) get generated.
3. Plugin-injected orderBy values — buildOrderByValues similarly merges extra enum values from the schema's orderBy type.
4. Transitive type resolution — generateCustomInputTypes previously only followed nested types ending with "Input", so enum types like VectorMetric (referenced by VectorNearbyInput.metric) were silently dropped. Now follows all non-scalar types present in the typeRegistry.

All changes are backwards-compatible — the typeRegistry parameter is optional throughout.

Tests added: 7 new test cases covering VectorFilter generation, plugin-injected condition fields, transitive enum resolution, field deduplication, plugin-injected orderBy values, and backwards compatibility. 14 snapshots updated (mechanical — VectorFilter added to always-generated scalar filter types).

Review & Testing Checklist for Human

• Broadened transitive resolution safety: The change in generateCustomInputTypes (line ~1619) now follows any non-scalar type that exists in the registry (not just *Input). The generatedTypes set prevents double-emission within this function, but types already emitted by earlier codegen stages (e.g., generateEnumTypes for table-column enums) could theoretically be duplicated in the output. Verify this doesn't cause issues with real schemas.
• Type representation for plugin-injected condition fields: Plugin-injected fields use typeRefToTs(field.type) while table-derived fields use scalarToTsType(fieldType) + ' | null'. Verify the nullability semantics are correct for both paths (table fields always get | null, plugin fields get nullability from the TypeRef wrapper).
• End-to-end validation: Run codegen against a real schema that uses graphile-pgvector-plugin and confirm the generated input-types.ts now includes embeddingNearby in condition types, VectorMetric enum, and distance-based orderBy values. The unit tests use mocked registries that may not perfectly match actual PostGraphile output.
• Suggested test plan: Deploy a local PostGraphile instance with pgvector + VectorSearchPlugin, export the schema, run pnpm codegen against it, and inspect the generated input-types.ts for completeness.

Notes

• Fixes constructive-io/constructive-planning#663
• Link to Devin session
• 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 and CI monitoring