You approach doesn't account for when you have a valid reason to insert the data again. Such would be the case when you got the original file with some or all erroneous data. So the question becomes should you delete at the source or at the destination, I think you should think this out.

I think you are fine with the PK clustered index on the fields you are joining on in this temp table. I don't think you need to cluster a separate id field, it just gets wiped at the next batch on not used.