cli: Refactor load_tables() to better match database

In preperaration for adding proper database support, let's refacotr
load_tables() to allow loading CSVs without needing to also generate the
"unique" columns. This will make it reusable for database code.

Additionally, let's convert timestamps and uuids at load time so that we
can better interface with a database. And since the database can't
handle NULL values, let's convert nan (what pandas fills NULLs from the
CSV with by default) with empty strings.

This is all temporary since the function will eventually be subsumed
into the upcomming datastore API.

Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>