feat(ui): multi-schema awareness across Editor, Brain, and Advisor

[venkateshsakamuri-lab]

Summary

End-to-end multi-schema UI pass so connections like Multi Schema Shop (crm / sales) no longer look like a single bare-name catalog.

P0 — Authoring

• Shared src/lib/schemaNames.js helpers (canonicalTableReference, objectKey, qualifyForSql, …)
• SQL Editor + Tables overview: schema-prefixed labels, qualified autocomplete/insertText, generated SQL (FROM crm.orders), collision-safe React keys, schema-aware search
• Clicking a table name inserts SELECT … FROM schema.table (chevron expands columns)
• Index/stats REST paths accept schema.table; Postgres/MySQL introspection scopes indexes by schema

P1 — Brain / Knowledge / ERD

• Schema Docs display + note persist use tableReference (no cross-schema note collisions)
• Knowledge @ / @@ suggestions show qualified labels; bare-name lookup only when unique
• ERD / classification / key-column grouping use qualified ids
• DetailsLibrary CSV example + NoteModal placeholder encourage schema.table

P2 — Advisor / ops copy

• Database advisor scans non-public schemas and emits schema-qualified CREATE INDEX … ON
• Performance cards / heatmap keep schema in labels
• Privileges accordion documents multi-schema GRANT pattern
• Dashboard-design skill nudges schema-qualified SQL

Verification

On Multi Schema Shop:

• Editor lists crm.customers, sales.order_items, sales.orders
• Click inserts FROM crm.customers
• Brain Schema Docs shows the same qualified names

Editor with schema-qualified tables
Generated SQL FROM crm.customers
Brain Schema Docs with crm/sales tables

Test plan

• Select Multi Schema Shop → Editor lists crm.* / sales.*
• Click table generates FROM schema.table SQL
• Brain Schema Docs shows schema.table
• Regression: Demo Shop (public-only) still shows bare table names
• node --test src/lib/schemaNames.test.js

Note: PR #53 is docs-only Cloud caveats; this PR is the multi-schema UI work.

_{To show artifacts inline, enable in settings.}

  

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[geekypunk]

Reviewed backend + the new shared util.

Backend. PostgresIntrospectionProvider.getTableIndexes now joins pg_namespace and filters with (?::text IS NULL OR n.nspname = ?) — fully parameterized, and every column is table-qualified per the repo's SQL rule. The null-schema branch keeps bare names working, so this is backward compatible. {tableName:.+} on the two /tables/{id}/… mappings is the right fix for schema-qualified ids, and PathPatternParser keeps it to a single path segment.

Shared util. schemaNames.js is pure, documented, and ships with tests. objectKey including the schema is what stops crm.orders and sales.orders colliding as React keys — the actual bug class this PR exists to close. qualifyForSql correctly leaves public/dbo bare.

One non-blocking note for a follow-up: with a bare table name the Postgres schema filter is disabled (schemaName stays null), so indexes can still merge across same-named tables in different schemas. That only matters for callers that haven't been migrated to encodeTablePathId; the UI paths in this PR all pass qualified ids. Worth tightening to the session search_path later rather than blocking here.