DR loses control of process on `_get_osfhandle`

Hi,

I'm running the latest DynamoRIO cronbuild, on Windows 10. I'm instrumenting a 64-bit Windows 10 application via `drrun`, using `-client_lib` to pass my DR client (which is built as a `dll`).

What I'm seeing *looks* like a loss of control, specifically on a call to `_get_osfhandle` in my client. My client and target application run fine together with the call stubbed out, but execution disappears (with no error messages) when it's re-added. WinDBG shows a normal exit, which makes me think that the target application is regaining control and terminating on its own.

Passing `-debug` to `drrun` produces the same behavior.

Let me know if logs would help -- I can try to clean some up.