cloud_job() should use most recent subdirectory of cloud.noindex

[hadley]

And cloud_check() should create this directory when it first runs.

I think this provides a simple way of automatically determining the correct job.