SPAs aren't completely legally binding. The FDA can do whatever it wants. I can see a Pazdurized CTGT taking apart the Cox regression analysis, i.e. using different stratification factors than the ones in the SPA in order to discredit the Cox. Then, even if the log rank p value is stat sig, the agency can claim that log rank is the secondary analysis in the SPA, and since the Cox regression primary analysis was flawed, DNDN needs to run another trial.
This is most likely unduly pessimistic, since I don't see CBER/CTGT losing purview of the BLA to CDER/OOD. However, this is how a negative scenario, even with stellar trial results, would play out if Pazdur were to succeed in wresting away the purview. Again, the more political pressure applied to the FDA by the advocacy groups, the less likely this scenario would happen, IMO.