thread init mcontext on Windows is the post-init state, not the initial state

#117 and 28058da are passing the CONTEXT of the resumption point for a new thread as the initial mcontext, which does not seem like what we want at all: the thread is going to run a lot of init code and we want the start of that code to be the initial mcontext. Nobody noticed this all this time so maybe it doesn't matter much. I hit it while adding #241 where I did not have LdrInitializeThunk as a kernel xfer, yet Dr. Memory was relying on that to mark the init stack properly as the thread init event's mcontext stack doesn't cut it.