This now respects the fetch_data/store_data pattern from the
loader-core.
So fetch_data is only one iteration on one svn revision. In that
call, we consume one svn revision as before and compute the
contents/directories/revision. The result is stored in internal
instance variable (as in pypi/debian loader for example).
And store_data, also in the same iteration, is in charge of passing
those data to the storage.
If the iteration is over, we build the snapshot and stops appropriately.
The internal computations did not change.
Note: I'm willing to rebase all this on @anlambert's current work to improve the loading speed (D433)