clustering step_down_p efficiency improvement?

this is admittedly an edge case, but when running TFCE and the first pass shows that every vertex is part of a significant cluster and should be excluded, there shouldn't even be a second pass, right? Admittedly it only took 1:05 for the second pass (versus 41:36 for the first pass) but honestly I find it odd that the code somehow managed to perform 10k permutations on zero vertices and spent a whole minute doing so.

stat_fun(H1): min=1.805931 max=14.442601
Running initial clustering
Using 73 thresholds from 0.00 to 14.40 for TFCE computation (h_power=2.00, e_power=0.50)
Found 20484 clusters
Permuting 9999 times...
100%|███████████████████████|  : 9999/9999 [41:36<00:00,    4.00it/s]
Computing cluster p-values
Step-down-in-jumps iteration #1 found 20484 clusters to exclude from subsequent iterations
Permuting 9999 times...
100%|███████████████████████|  : 9999/9999 [01:05<00:00,  153.38it/s]
Computing cluster p-values
Step-down-in-jumps iteration #2 found 0 additional clusters to exclude from subsequent iterations
Done.

Start by locating the TFCE step-down-in-jumps implementation associated with step_down_p and reproduce the behavior shown in the issue. The work is done when a run with all vertices excluded does not perform the additional 9,999-permutation pass, while the reported clustering and p-value results remain correct.