DuckDB read_json_auto returns BIGINT as JS bigint, breaking Zod and JSON.stringify
Importing JSONL with @duckdb/node-api's read_json_auto infers every JSON integer as BIGINT, and getRowObjectsJS() hands those back as JS bigint. That silently fails Zod z.number()/z.number().int() validation and makes JSON.stringify throw 'Do not know how to serialize a BigInt' the first time a route handler returns the row - and no count-based test catches it, because count(*) tests never read a value. Cast every integer column explicitly in the CREATE TABLE AS projection, but do not blanket-cast to ::INT: ids are fine, while money is not, since a 39-billion value raises 'Conversion Error: out of range for the destination type INT32'. Use ::INT for ids and ::DOUBLE for money/prices, and remember aggregates too - count(*)::INT, since even 'literal - count(*)' can come back BIGINT. Related trap: DuckDB 2.x deprecated the '->' lambda arrow, so list_filter(col, x -> x.k = 1) now errors and must be list_filter(col, lambda x: x.k = 1).