Whoa!!!
First, you never want to export data from a gdb to a shapefile and then re-import it back into a feature class. Doing this could compromise the topological integrity of the data. If you need to copy & paste (import/export) data from a gdb then you should keep it in a gdb for as long as possible. There are things like precision and tolerance that behave differently in a feature class vs a shapefile.
Second, are you using the specified topology rules for migrating to the fabric? The "must not intersect..." is referring to places where parcel lines are not broken at their intersections. There are 2 ways to handle these: use Planarize on the parcel lines or shift+select in the Error Inspector, Right-click, choose "split." If the errors persist after this, then it is possible that you get the error due to a self-overlapping line. I know this error pops-up on an basic overlapping line even though it wants you to subtract a line.
You may have come across mention of "marking errors as exceptions" which can be loaded into the fabric, but you only want to use this when they are legitimate. It's hard for me to think of any examples b/c they've done such a nice job with the model but then every organization is different.
Furthermore, it will be a headache to clean up these errors (sliver poly's, holes, etc) in the fabric after the fact. The concept of the fabric is that the parcels are highly dependent on the COGO so you will want your data to improve as you move forward--not move forward and then go back fudge some things to close, and then finally add COGO as you progress.
Can you possibly provide some more info on the number and type of topology errors you're getting? Once you've edited the data using a topology and saved them, the changes are made to the data. So deleting it will not affect your data, but you will need it for fabric migration.