Refactor integration tests to have a single point of entry

[arostamianfar]

We currently have 3 integration test "binaries". They have several common elements (the logic is mostly shared in the run_tests_common.py lib). It's cleaner to have a single binary where the flags specify which integration tests to run and it figures out the appropriate sub-library to run (we likely need a common interface for defining the tests as well).