Hey ,
Unfortunately based on my understanding, in case some activity has failed within for activity and you rerun from failed activity, it would again iterate for the entire number of iterations and not just the failed one.
Now, in case if you want to run just the failed iteration, then you would have to keep a log table and have a flag updated at the end of every iteration for a successful one.
AND then run the entire pipeline again as it would now take the iteration of just failed one (the one where flag was not updated)
You can raise a feedback for #1 with MSFT