[spark] Fix a shifted timestamp written from before a zone left local mean time

[cxzl25]

Purpose

SET spark.sql.datetime.java8API.enabled = false;
SET spark.sql.session.timeZone = Asia/Shanghai;

CREATE TABLE t (id INT, ts TIMESTAMP) USING paimon;
INSERT INTO t VALUES (1, timestamp'1900-01-01 00:00:00');
SELECT cast(ts as string) FROM t;

-- before this change: 1900-01-01 00:05:43
-- with this change:   1900-01-01 00:00:00

Tests

Add UT