SQLiteRows.Columns() returns outdated state

This is a little difficult to explain so I apologize if it seems confusing.

The problem is that when there 2+ live connections and 1 of the connection changes the schema, calling rows.Columns() from any of the other connections return an outdated list of columns (aka. the state before the schema change).

Below is a minimal reproducible:

package main

import (
	"database/sql"
	"log"
	"os"

	_ "github.com/mattn/go-sqlite3"
	// note: uncomment to test with the pure go port (the driver name is "sqlite")
	// _ "modernc.org/sqlite"
)

func main() {
	os.Remove("./foo.db")

	db1, err := sql.Open("sqlite3", "./foo.db")
	if err != nil {
		log.Fatal(err)
	}
	defer db1.Close()

	db2, err := sql.Open("sqlite3", "./foo.db")
	if err != nil {
		log.Fatal(err)
	}
	defer db2.Close()

	// ensure that there is a live db1 connection
	if err = db1.Ping(); err != nil {
		log.Println("failed to ensure db1 live connection", err)
		return
	}

	// create the dummy schema
	_, err = db1.Exec("CREATE TABLE foo (id INTEGER PRIMARY KEY, title TEXT)")
	if err != nil {
		log.Println("create table failure:", err)
		return
	}

	// ensure that there is a live db2 connection AFTER the table creation
	if err = db2.Ping(); err != nil {
		log.Println("failed to ensure db2 live connection", err)
		return
	}

	// rename for example the column through one of the connections
	_, err = db1.Exec("ALTER TABLE foo RENAME COLUMN title TO title_new")
	if err != nil {
		log.Println("column rename failure:", err)
		return
	}

	// apparently running PRAGMA quick_check, optimize, etc. seems to fix the sync state
	// db2.Exec("PRAGMA quick_check")

	rows, err := db2.Query("select * from foo")
	if err != nil {
		log.Println("select failure:", err)
		return
	}
	defer rows.Close()

	columns, _ := rows.Columns()

	log.Println("Expected [id, title_new], got", columns)
}

Run the minimal Go reproduction using the go-sqlite3 driver, then start at database/sql Query and Rows.Columns behavior and trace the driver's handling of schema changes across live connections. Done means the reproduced query returns [id, title_new] from the second connection after the first connection renames the column, with regression coverage for that scenario.