Inconsistent behavior after "400 Bad Request"

The issue likely affects other error codes.

After the server returns 400 Bad Request, it does not close the socket, and also it does not allow to reuse the same socket. This leads to unexpected behaviour when the library tries to use the same socket next time.

The TCP capture for iteraction is here: cbas-bad-request-issue.pcapng. For comparison, here is how N1QL behaves, where 400 Bad Request does not invalidate the connection (IMHO the CBAS could behave in the same way: n1ql-bad-request.pcapng.