if on a remote server (e.g. using `auth_local_webserver=False`), then don't cache user credentials by default

[tswast]

If on a remote server (e.g. using auth_local_webserver=False), then there's a good chance the code is running on a shared machine (e.g. a VM or Colab notebook). In these cases, we should not cache credentials to a local filesystem, as there is a chance that other users have access to the same machine.

If I remember correctly, Colab already has patched this library to disable the caching when installed there, but it'd be great if we could do this by default in a way that works for all potentially-shared environments.

[tswast]

Note: in addition to tackling this here for pandas-gbq, we should update the defaults in pydata-google-auth for folks using that library directly. pydata/pydata-google-auth#63