parse_test_failure.py doesn't handle errors outside of test cases

The parse_test_failure.py script generates JUnit-style XML report files for tests that either failed to generate one themselves, or whose generated XML file is missing critical information. For example, gtest's XML report generation logic isn't aware of TSAN warnings, so a test with a TSAN warning ends up with a clean XML report, even though the report ought to show that some test failed with a TSAN warning.

Unfortunately, parse_test_failure.py silently drops e.g. TSAN warnings that were reported outside of a particular test run. This can lead to missing test failures in test dashboards and the like.