Clean shared state pollution to avoid flaky tests

What is the purpose of this change

This PR is to clean shared state pollution between tests:

Test1: org.apache.hadoop.mapreduce.v2.app.TestJobEndNotifier.testNotificationOnLastRetryShutdownWithRuntimeException
Test2: org.apache.hadoop.mapreduce.v2.app.TestMRAppMaster.testMRAppMasterShutDownJob

Test1 can pollute the shared state with Test2, which can lead Test2 to fail after running Test1. It may be better to clean state pollutions so that some other tests won't fail in the future due to the shared state polluted by this test.

Reproduce test failure

First run Test1, then run Test2 in the same JVM

Expected result

The tests should run successfully when multiple tests that use this state are run in the same JVM.

Actual result

Run Test1 and Test2, Test2 fails:

[ERROR] TestMRAppMaster.testMRAppMasterShutDownJob:530 Expected shutDownJob to exit with status code of 0. expected:<0> but was:<1>

Fix

Reset FirstExitException at the start of Test2 to clean the pollution from Test1.