Data Connections: Column and table completions

Context

Data Connections already has the schema for a db at the typescript layer which includes a tree with tables, columns, viewws independent of the R/Python session. Can we surface this information as part of SQL autocompletions?

Scope

1. Wire table, column, view completions
2. Scope column completions to tables actually referenced in the current query where feasible, rather than suggesting every column in the connection
3. Handle large catalogs: progressive disclosure or lazy loading for connections like Snowflake/Redshift where the full tree could be large enough to matter for editor responsiveness
4. No active connection should have no schema suggestions
5. Multiple active connections: decide how completions scope (active/focused connection only vs. some kind of picker)