Description
Currently, for the various fault injection crash points, we exit using LOG(FATAL). This has two issues:
1) we end up with a fair bit of log spew that isn't particularly useful, and which also confuses the "Diagnose" button on the flaky dashboard
2) our tests which are typically waiting for the server to crash can't distinguish the "expected" crash (injected fault) from any other crash (eg a bug that causes a SEGV or CHECK failure)
To improve this, we can make the fault injection exit with some special error code, and have the tests that wait for the crash fail the test if they see a crash with another exit code (indicating SEGV or ABRT)