Good questions. Some of this is covered in the tutorial on fermions if you need to read more about it: http://itensor.org/docs.cgi?page=tutorials/fermions
1. Yes, you do have to explicitly put in F operators when measuring correlation functions still. The reason is that when you are getting operators from the sites.op method, it is returning individual tensors to you and currently these tensors have no way of knowing about the existence of fermions on other sites since they are just local tensors. (In the future I plan to experiment with making fermionic IQIndices automatically anticommute - stay tuned.)
2. You should not put four fermion operators on the same site anyway, since this would be zero by definition when acting on a local Hilbert space that can only have at most one spin up and one spin down electron.
3. Yes, by respecting the fermion anti-commutation I meant writing the order and the signs of the terms you put into the AutoMPO in the same way you would on paper when considering the correct fermion anticommutation rules. So there is more than one correct way to input the terms, but you just have to make sure that whatever choice you make matches what you would write on paper (e.g. for the Hermitian conjugate of the hopping term Cdag_i C_j you could either put + Cdag_j C_i or -C_i Cdag_j (here assuming i != j).
Here is a snippet of the code I am currently using to input the two-particle terms of a chemistry Hamiltonian, where "V" means a particular V_ijkl:
if(i != j && k != l)
ampo += V,"Cdagup",i,"Cdagup",j,"Cup",k,"Cup",l;
ampo += V,"Cdagdn",i,"Cdagdn",j,"Cdn",k,"Cdn",l;
ampo += V,"Cdagup",i,"Cdagdn",j,"Cdn",k,"Cup",l;
ampo += V,"Cdagdn",i,"Cdagup",j,"Cup",k,"Cdn",l;