fix(codegen): detect pgvector/tsvector via GQL types; use _meta for blueprints

[pyramation]

Summary

Two related simplifications that reduce reliance on direct Postgres introspection:

1. Codegen: detect pgvector/tsvector via GQL type names, not pgType

PostGraphile's codec plugins already translate PostgreSQL types into distinct GraphQL scalars — vector → Vector, tsvector → FullText, geometry/geography → GeoJSON/etc. The codegen can detect these directly from the introspection schema without any additional metadata.

Changes:

• isEmbeddingField(): Now checks gqlType === 'Vector' (primary) instead of pgType === 'vector' (which never fired). Retains the legacy name.endsWith('embedding') && Float[] heuristic as fallback.
• isTsvectorField(): Added gqlType === 'FullText' fallback alongside the existing pgType === 'tsvector' check.
• buildSearchHandler(): Tsvector detection in the search CLI handler now uses gqlType === 'FullText' instead of pgType === 'tsvector'.

What was removed (added earlier in this branch, never shipped):
The branch previously introduced metaFile config, --dump-meta CLI, enrichPgTypesFromMeta, MetaFieldInfo/MetaFieldType types, and associated test files (~630 lines). All removed once we realized GQL type detection makes the entire pgType enrichment machinery unnecessary.

M:N junction key enrichment from _meta remains unchanged — that genuinely requires _meta and is unaffected by this PR.

2. Node-type-registry: replace --introspection with --meta for blueprint type generation

Customers without direct Postgres access can now use the _meta GraphQL query output to generate typed blueprint interfaces. The --meta flag accepts _meta.json in any of these formats:

• Raw MetaTableInfo[] array
• { tables: [...] } (MetaSchema wrapper)
• { data: { _meta: { tables: [...] } } } (full GQL query result)

When --meta is not provided, static fallback types are used — no breakage.

Types renamed from IntrospectionFieldMeta/IntrospectionTableMeta to MetaFieldInfo/MetaTableInfo to align with MetaSchemaPlugin naming.

Review & Testing Checklist for Human

• FullText scalar assumption: Verify that no PostgreSQL types other than tsvector map to the FullText GraphQL scalar in PostGraphile/TsvectorCodecPlugin. If another type does, isTsvectorField() would produce false positives.
• Vector scalar requires VectorCodecPlugin: The gqlType === 'Vector' check only works when VectorCodecPlugin is active. Without it, pgvector columns are invisible in the schema entirely (silently dropped), so this should be fine — but worth confirming.
• --introspection → --meta is a breaking flag rename: Verify no CI scripts or docs reference --introspection for generate-types.ts. The package.json script (pnpm generate:types) does not pass it, so this should be safe.
• No runtime validation on --meta JSON: The parsed JSON is cast to MetaTableInfo[] without schema validation. Malformed input falls through to static fallbacks with a console warning, which is graceful but could silently produce incomplete types if the JSON is almost-but-not-quite right.
• Test manually: Run codegen against a database with pgvector, tsvector, and PostGIS columns. Verify the generated docs/CLI correctly categorize these fields (embedding, search, geospatial groups) without any _meta or metaFile configuration. Also test pnpm generate:types with and without --meta to confirm both paths produce valid output.

Notes

• isPostGISField() retains its existing pgType check alongside GQL type pattern matching. The pgType path is effectively dead code (never populated from introspection), but harmless — it would fire if pgType is ever set by another mechanism in the future.
• Same applies to isTsvectorField() which still has pgType === 'tsvector' as a first check before the FullText fallback.
• The node-type-registry package has no unit tests (pnpm test exits with "no tests found"). The build passing is the primary verification for the generate-types.ts changes.

Link to Devin session: https://app.devin.ai/sessions/c92c3a11450342f8875625a60fa1be28
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