…places, but without calling wb_close_socket. When the connection was finally torn down, wb_close_client would only close the socket when WB_ALIVE was still set. Seems more robust to check socket != WB_INVALID_SOCKET?
…_update_client. (Seen when specifying content-length and WB_CLOSE_AFTER_RESPONSE was set)
… for just N-1 connections is reserved. Reading into the first connection header_buf was corrupting srv->connections. Wouldn't cause any problems until 4 connections came in simultaneously