Timestamp which dtype is Float64 are treated with timezone

[zhouwuji]

Describe the bug
When I insert a df with a timestamp into a table, if the dtype of that timestamp column is Float64, it is considered to be local_time, and then the value inserted into the clickhouse produces a wrong offset, which is offset for me by 8 hours.

To Reproduce

from clickhouse_driver import Client
import pandas as pd

port = 9000
client = Client(
    host='192.168.1.160',
    settings={
        'use_numpy': True},
    port=9000,
    user='default',
    password='')

sql1 = '''

drop table if exists dt;
'''
sql2 = '''
CREATE TABLE dt
(
    `timestamp` DateTime64(9),
    `event_id` UInt8
)
ENGINE = TinyLog;
'''
client.execute(sql1)
client.execute(sql2)
data = [
    (1659099540588999936, 1),
    (1659099540588999936, 2),
    # ('2022-07-29 20:59:00.588999936', 3),
    (1659099540588999936.0, 2),

]

df = pd.DataFrame(data)
df.columns= ['timestamp', 'event_id']
df.info()
res = client.insert_dataframe(f'INSERT INTO dt VALUES', df)

df2 = client.query_dataframe('SELECT *,toUnixTimestamp64Nano(timestamp) FROM dt;')
print(df2)

output:
                   timestamp  event_id  toUnixTimestamp64Nano_timestamp_
0 2022-07-29 12:59:00.588998         1               1659070740588998912
1 2022-07-29 12:59:00.588998         2               1659070740588998912
2 2022-07-29 12:59:00.588998         2               1659070740588998912

Expected behavior
I hope the value of timestamp do not minus 8hours.

Apparently, it treated my UTC time as UTC+8

                   timestamp  event_id  toUnixTimestamp64Nano_timestamp_
0 2022-07-29 20:59:00.588999         1               1659099540588999936
1 2022-07-29 20:59:00.588999         2               1659099540588999936
2 2022-07-29 20:59:00.588999         2               1659099540588999936

(1659099540588999936-1659070740588998912)/1e9/3600
Out[28]: 8.000000000284444

Versions

• numpy==1.19.0
• pandas==1.0.5
• clickhouse-driver==0.2.2
• ClickHouse server version==21.8.4.51
• Python version==3.7