Use global stash instance instead of local cluster instance

This seems like something that should be global so login
attempts are accumulated across all data centers

At first I thought this might be related to T173888,
but now I don't think so, as logins would be POSTs
so should all be in the master db anyways.

Change-Id: I8d64a8b1c6607cb5b32a25182ec3496477361fdd