Metrics for tracking query failure reasons

A query can fail for various reasons when run in Impala:

• analysis failure (e.g. SQL syntax mistake)
• access right issue (e.g. no privilege for certain ops on a table)
• memory limit exceeded
• scratch limit exceeded
• EXEC_TIME_LIMIT_S exceeded
• other guard rail exceeded
• executor crashes
• rpc failures
• corrupted data files
• client disconnected or session timed out

This JIRA tracks the effort to explicitly classify query failures into some high level categories and expose them in the metrics. This should provide us a clearer view of what types of failures contribute the most to bad user experience for a given cluster.