Timeout for metadata loading from an impalad

Today, an impalad requesting the metadata of a table to be loaded from the catalogd will wait for 2m and then keep retrying every 2m by sending an RPC to the catalogd. However, there is no upper bound on the total wait time, so the impalad will retry indefinitely. If something goes wrong on the catalogd side, such a query could hang and users won't be able to cancel it (see IMPALA-915).

We should consider adding a timeout or another solution to avoid hangs in such pathological situations.