Better handling for master's response (0xff) to COM_STMT_PREPARE

When prepared statements are routed to both master and slave by readwritesplit router, in some cases we may get warnings like this:

2019-12-18 14:07:33 warning: (62418270) Slave server 'db2': response (0x00) differs from master's response (0xff) to COM_STMT_PREPARE: `SELECT ... FROM ... WHERE (`t1`.`c1` = ?) AND ...)`. Closing slave connection due to inconsistent session state.

I suspect it may be cause by the need to re-prepare the query on master when/if ALTER was executed there.

In any case, better logging for the reasons of response 0xff from master would help.