DuckDB: host integration and FileQueryEngine

[anitnilay20]

Context

Core infrastructure for the DuckDB file query engine epic. All other sub-issues (filtering #53, sorting #54, aggregation #55, query editor UI) depend on this.

What to build

A FileQueryEngine struct owned by the host, one instance per open file tab.

Format detection

On file open, detect whether DuckDB can read the format natively:

Format	DuckDB function
JSON / NDJSON	read_json_auto('path')
CSV / TSV	read_csv_auto('path')
Parquet	read_parquet('path')
Arrow IPC	read_arrow('path')
Excel	st_read('path') (spatial ext)
Other	fallback: get-range ingestion

For native formats, register a view: CREATE VIEW data AS SELECT * FROM read_json_auto('/path/to/file.json'). DuckDB scans lazily — no full load upfront.

For plugin-loaded formats, bulk-pull via get-range and insert into an in-memory table. Show a progress indicator during ingestion; do this async off the UI thread.

FileQueryEngine API

pub struct FileQueryEngine { /* DuckDB connection, ingestion state */ }

impl FileQueryEngine {
    pub fn open_native(path: &Path, format: NativeFormat) -> Result<Self>;
    pub fn ingest_from_plugin(records: Vec<String>) -> Result<Self>;
    pub fn query(&self, sql: &str) -> Result<QueryResult>;
    pub fn schema(&self) -> Result<Vec<ColumnInfo>>; // for editor autocomplete
}

Changes

• Add duckdb crate to Cargo.toml (feature flags: bundled for static linking)
• src/query/engine.rs — FileQueryEngine
• src/query/mod.rs — format detection, ingestion coordinator
• Host file-open path: after plugin open(), attempt native DuckDB open; fall back to plugin ingestion

Acceptance criteria

• DuckDB opens JSON, CSV, Parquet files directly without plugin ingestion
• Plugin-loaded formats ingest via get-range into an in-memory table async
• query(sql) returns typed rows (not stringified)
• schema() returns column names and DuckDB-inferred types
• One DuckDB connection per tab; connection dropped when tab closes
• No UI freeze during ingestion (async worker)