Since the point of the Cache is to own the directories and all the
metadata about requirements, they need to own this stuff.

Currently we delete succesfully installed requirements and not ones
that failed. Then (most of the time) we delete the requirements cache
that contained them, deleting everything. This means that useful
debugging context is lost (when we don't delete the cache), and we've
got complexity serving no point. Instead, either keep everything or
delete it all.

Its useful to know whats happening, so push the existing logging down
to debug, and store the skip reason (which will be constant for a
given req) on the req itself.