Regression in copy records performance between 0.30.0 and 0.31.0

When run on different versions (0.30.0 vs 0.31.0), the following script reproduces the issue and consistently shows a performance regression of approximately 22%:

import asyncio
import asyncpg
import time

COUNT = 5_000

async def run():
    conn = await asyncpg.connect("postgresql://user:password@localhost/postgres")

    await conn.execute("CREATE TABLE IF NOT EXISTS users (id SERIAL PRIMARY KEY, username TEXT, email TEXT, age INT)")

    RECORDS = [("John Doe", "john.doe@example.com", 42)] * 10_000

    start = time.perf_counter()
    for _ in range(COUNT):
        await conn.copy_records_to_table("users",  records=RECORDS,  columns=["username", "email", "age"])
    end = time.perf_counter()

    await conn.close()

    return end - start


if __name__ == "__main__":
    total = asyncio.run(run())
    print(f"Total time: {total:.5f} seconds")
    print(f"Average time per loop: {total / COUNT:.8f} seconds")

(common) [e.lee@OC0000842]$ pip install asyncpg==0.30.0
(common) [e.lee@OC0000842]$ python -m example
Total time: 89.15468 seconds
Average time per loop: 0.01783094 seconds

(common) [e.lee@OC0000842]$ pip install asyncpg==0.31.0
(common) [e.lee@OC0000842]$ python -m example
Total time: 108.57080 seconds
Average time per loop: 0.02171416 seconds

Inizia eseguendo lo script asyncio fornito con asyncpg 0.30.0 e 0.31.0 su PostgreSQL, confermando la differenza di tempo segnalata per copy_records_to_table. Segui il percorso di copy_records_to_table per identificare cosa è cambiato tra queste versioni; il lavoro è completo quando la regressione è spiegata e il benchmark non mostra più il rallentamento segnalato.