As discussed in #214, source_created and source_changed didn't really
have clear definitions. In truth, their names and separation betray
their origins as model receivers in earlier versions. The "source group"
abstraction helped us get away from thinking about things exclusively in
terms of models, but these remained as an artifact.
Differentiating between when the generated file content is required and
when the generated file is just required to exist gives us more
flexibility with strategies.
Reading from an `ImageCacheFile`, will result in accessing its `file`
attribute repeatedly which would result in the `before_access` signal
being dispatched, which in turn would result in many unnecessary calls
to the image cache backend. With this change, we don't send
`before_access` if the file has already been created.
Similarly, we don't need to try to generate the image if we know for
certain that it's already been generated (because we have a reference
to it).
Changed my mind about 04aa72c1f9. It's
just a better description, even if different strategies can change the
behavior so it isn't really very cache-like.
