segfault during lcb handle-level error handler

This is related to CCBC-120 (stack trace there). Will fix accordingly soon.

Basically the error handler checks for the 'ctx' which is not reset in-between calls.

The ideal solution should be for start_loop to accept a cookie argument and call lcb_set_cookie on it. This way there is always a valid cookie.

Otherwise the cookie remains a stale pointer and when the global error handler is called, it crashes.