[source storage] refactor source sql generation and results handling to allow repository side callbacks

for instance with the BytesFileSystemStorage, before this change:
* fspath, _fsopen function were stored procedures executed on the database
  -> files had to be available both on the repository *and* the database host
* we needed implementation for each handled database

Now, those function are python callbacks executed when necessary on the
repository side, on data comming from the database.

The litle cons are:
* you can't do anymore restriction on mapped attributes
* you can't write queries which will return in the same rset column
  some mapped attributes (or not mapped the same way) / some not

This seems much acceptable since:
* it's much more easy to handle when you start having the db on another host
  than the repo
* BFSS works seemlessly on any backend now
* you don't bother that much about the cons (at least in the bfss case):
  you usually don't do any restriction on Bytes...

Bonus points: BFSS is more efficient (no queries under the cover as it
was done in the registered procedure) and we have a much nicer/efficient
fspath implementation.

IMO, that rocks :D