Off-by-one in decoy selection excludes the real output from the first RPC batch

The intended first-iteration guard is keyed on `iters == 0` but `iters` is incremented before the check, so the real output is never mixed into the first `get_unlocked_outputs` query; an on-path node computes the set difference between the on-chain ring and that first candidate set to isolate the true spend, collapsing the effective anonymity set to 1.