Before, no particular treatment was done on the user attributes.
This caused some issues because python-ldap systematically returns
the attribute value as a list (even if it's mono-valuated).

Now we recover the attributes used in the group attr templates,
and we "normalize" the user attributes before using it in add_to_groups
and del_from_groups.

By normalize, we mean, transforming the list to it's unique value.
In case the attribute doesn't exist or is multi-valuated, it raises an
error.