concurrency bug in ThreadReferenceTest.test_resume

[csuter]

test is currently flaky. still investigating.

possibly has something to do with

• initialization order of main thread vs child thread (see ThreadReferenceTest definition for test class detail)
• more remotely: javac optimization of test java code. the code basically does nothing (sleeps a couple threads), so maybe the compiler is eliminating code. adding println's affects behavior, somewhat in support of this theory. But it's much more likely the println's affect timing.