Stop allocating and deallocating memory as part of the test itself.
There's still the fopen, but since our stdio reuses existing structs,
that doesn't seem to be a problem in practice.
Bug: http://b/67077411
Test: ran tests with --gtest_repeat=1000
Change-Id: I99de5de0911161ec04afe75653075f1ccefb01a5
In pthread_leak.detach test, compare the memory usage after two same
scenarios - all children threads have exited. Thus, test emulated by
NativeBridge, which may reserve a memory pool, will pass if it's not
unbounded leak.
Bug: https://issuetracker.google.com/37920774
Test: CtsBionicTestCases
Change-Id: Ia51aa67bacb588284747652b36ea5d8e85bea832
