update documentation or tutorial to avoid the mistake that shangsi made

[jovo]

i don't recall what it is anymore, so i don't know how to do this, and @cshen6 seemed to be uninterested in doing it. so i'm closing it. @cshen6 re-open if you know how to fix this issue.

[cshen6]

I still remember what it is; and related to cep's post-model question, i.e., one should use sample mgc stat, rather than individual local corr for testing on real data.

I added comments in the main permutation test; hopefully it will avoid potential mis-use.