Bulk Operation

TEZ-803 Remove implicit assumption that task outputs are on the same node as the one it ran on

TEZ-803 Bail out early if a vertex has too many failures

TEZ-803 Read failure diagnostics should be available to the user

TEZ-803 Factor host failures into Shuffle failure heuristics

TEZ-803 Output verification should be strict enough to disregard relaunched attempt.

TEZ-803 Add unit tests for DAG with broadcast

TEZ-803 Ensure appropriate number of data movement events are generated by a task