I can do a cell like this

-- ``sql matches ..
select * 
from "http://localhost:5000/outputs.without_replacement.parquet"

and it works, but it's incredibly inefficient. I see the parquet file being fetched several time for an single cell.
Later if I reference ${matches} it just reinvoke the select from http.. and makes more requests.

Could we have a special sql block to execute on the main duckdb connection ? So that I can

create table matches as 
select * 
from "http://localhost:5000/outputs.without_replacement.parquet"