I was using the ODBC driver and found that the problem occurred in a temporary table.
However, when I worked with C# with Npgsql (PostgreSQL .NET driver), there were no problems even though connection pooling was working. When a connection is reused internally, DISCARD ALL is done first. So the temporary table is also automatically dropped.
I also tested MySQL, and the MySQL ODBC driver worked without problem. However, looking at the network protocol, it seems like the authentication process is almost done again rather than a connection reset. The process took quite a long time because it seemed that only the TCP connection was being reused rather than FIN. What's interesting is that the ODBC data source manager is installed with pooling turned off by default.
The default setting of MS SQL Server's ODBC driver is to use pooling, and it works well with or without pooling. Internally, sp_reset_connection is executed to initialize reused connections.