fix: drop pool_pre_ping, it is unusable on aiomysql

[manueltorres0]

Every ETL job has failed since the last deploy — 8 of 8 upserts plus the dimensional refresh, which had never failed before:

sqlalchemy/dialects/mysql/pymysql.py, in do_ping
dbapi_connection.ping()
TypeError: AsyncAdapt_aiomysql_connection.ping() missing 1 required
positional argument: 'reconnect'

SQLAlchemy's pymysql dialect chooses its ping call by inspecting PyMySQL's signature: with reconnect defaulting to True it sends ping(False), otherwise ping(). The aiomysql async adapter declares ping(self, reconnect) with no default, so the second form always raises — and every pooled checkout goes through it once pool_pre_ping is on. The first connection of each job succeeds (new connection, not pinged) and the reused one dies, which is why jobs failed ~4s in.

PyMySQL was unpinned (transitive via aiomysql), so it resolved to 1.1.2 locally and 1.2.0 in the image, and 1.2.0 is where the reconnect default flipped to False. That is why this passed local verification and broke in production; the check should have run in the image, and now does.

Replaces it with pool_recycle=600, which addresses the original problem — a connection dropped while idle being reused ~12 minutes later — without touching the ping path, and works under both PyMySQL versions. Verified in python:3.12-slim against MySQL:

PyMySQL 1.2.0 pool_pre_ping=True -> TypeError (reproduces prod)
PyMySQL 1.2.0 pool_recycle=600 -> OK
PyMySQL 1.1.2 pool_recycle=600 -> OK

PyMySQL is pinned as well, so local and the image stop diverging.