Data source: DuckDB plugin

Embedded analytical SQL engine (local files / Parquet). **SQL**.

Builds on the **Seshat** foundation (#93, shipped in #99): host `tcp-client` (+ host-side TLS), owner-thread query execution, `secure-storage` for credentials, and the `UiNode` DSL (connection dialog, dock tabs, results grid, `json-tree`).

## Constraint (key risk)
- DuckDB is a **library, not a server** — there's no socket to talk to. Under `wasm32-wasip1` we can't just open a TCP connection. Options:
1. Compile DuckDB to wasm and link it into the plugin (heavy), or
2. the **subprocess plugin path** (`docs/DATABASE_PLUGINS.md` Option C) running a native DuckDB.
- Opens a `.duckdb`/Parquet/CSV file via the existing `file-access` world.

## Reuse
- `DbAdapter`, results grid, schema browser (once an execution path exists).

## New work
- Decide + implement the embed/subprocess execution path; file open via the file dialog.

---
Part of the data-source plugins epic #109.

Start with docs/DATABASE_PLUGINS.md, especially subprocess plugin path Option C, then inspect the Seshat foundation from issues #93 and #99 and the existing file-access and DbAdapter paths. Decide whether the DuckDB execution path is embedded or subprocess-based, then verify that a .duckdb, Parquet, or CSV file can be opened and queried through the existing results and schema UI.

Escrita pelo modelo de indexação a partir do texto da issue.