Exadata or not, Partitioning and Parallel DML is the way to do that automatically.
What I understand is that you needed your metadata table in order to add the 'partition' clause to the insert clause in order to lock only the partition when you insert in direct path. is that right ? Then Parallel DML should do that automatically. Well I've done that kind of insert as select in parallel both in non exadata RAC and in Exadata, and it scales well.
Exadata will add the startscan feature for the SELECT part, and that's usually very good as you don't need all the rows/columns.
Question 2 is also not related to Exadata, not even RAC. If you're doing parallel insert, then each parallel process will insert in its own segment (partition).
About auto DOP, take care on Exadata because parallel_degree_policy=AUTO enables also in-memory parallel query, which you may not want on Exadata.