The thunk placement algorithm is better (lower cost, better results) when stubs immediatelly precedes text.
Also, when stubs comes first AND PGO/FDO places hot code at the front of text, then stubs are reachable without thunks from low __text addresses.
The test cases now work with either layout.