You should be happy that you have your data warehouse. All your data is already sourced, a great place to start with dimensional design.

To your questions:

SSAS shines at doing calculations on the fly, your grain should be lowest possible, one line per article per invoice, your measures likely to be Quantity and Sales Price.

Typically related measures reside in the same measure group, and typically each measure group has its own fact table. You could denormalize the fact table, however your ETL gets complex and you potentially lose history.

Personally, I prefer simple ETL, as you mention, complex ETL is hard to maintain. Generally, best practice today is to have one package per subject area.

In your scenario:

one package per dimension

one package per measure group

I assume the “using SSAS features” is cube processing, they can go in one package

As a nod to Koen, some book references: http://www.bidwbooks.com/learn-dimensional-modeling-basics-to-advanced/

You should watch this http://sqlbits.com/Sessions/Event11/Data_Modeling_for_Analysis_Services_Cubes