Optimize DuckDBClient’s handling of small Parquet files

[mbostock]

We currently use CREATE VIEW for Parquet files:

`CREATE VIEW '${name}' AS SELECT * FROM parquet_scan('${file.name}')`

Because views are not physically materialized, this can cause many range requests as DuckDB reads the file. Sometimes the overhead of these range requests outweighs the benefits of not loading the entire file.

Using the new file.size property #1608, we could employ a heuristic that switches to CREATE TABLE instead of CREATE VIEW for “small” Parquet files. We might also want an option to provide an explicit hint to use a materialized table instead of a view.