Hey @tahseen,
I think either approach would work. One advantage to joining the data prior to loading is the elimination of the extra time to stitch edges (and possibly introduce errors).
For a terabyte of data, you’ll definitely want to use the Bulk Loader. Have a look at the vlg repo, specifically the section on data loading: vlg/notes/3. Data Importing.md at main · dgraph-io/vlg · GitHub. And there may be other things in this repo you might find useful as you attempt your import.