Add dot-notation to policy language

Currently, policy authors must write policy over a flattened-table
representation of data sources, even if those data sources natively
represent their information using structured objects, e.g. dictionaries
of lists of dictionaries. And someone must write drivers that transform
the information exposed natively by data sources into flat tables.

This blueprint aims to give policy writers the ability to write policy over
structured objects and to eliminate the need to write drivers that
flatten the structured objects stored natively by data sources into tables.
At least, for most cases we could make these improvements in usability, but
we will retain the current infrastructure to handle those cases where
some manually written data source transformation is necessary.