As we call AttachCurrentThread() in the constructor, it seems logical to call
DetachCurrentThread() in the destructor. Some warning messages in the logcat
indicated that DetachCurrentThread() hadn't been called for a thread that
died.

Actually I would prefer to call both AttachCurrentThread() and
DetachCurrentThread() in the actual thread function, instead of somewhat
haphazardly and imlicitly in the AndroidSalInstance constructor and
destructor. Do we know for sure that the lifetime of the AndroidSalInstance
singleton (is is a singleton, right?) is 1:1 coupled to that of the LO "main
thread"?

Change-Id: Ifcc0e0b9af88b19389c416c5646499d44ad0e941