My opponent goes on bf with heron, he doesnt have played card yet. I use back off, he use defy, i use defy. Chain resolve, my defy works on his defy. He want to react in the chain with not so fast played for free. Does that work even if his defy doesnt resolve ?

Partly — he can play Not So Fast in that window, but no, it won't be free.

The play itself is legal. The chain resolves one item at a time, so once your Defy counters his Defy and clears it, a fresh priority window opens before Back Off resolves, and he can add a Reaction then (Rule 340.4; FAQ #9035, #9457). Not So Fast is [Reaction], so it's legally timed even in the closed state, and Back Off is a valid target because it chooses his friendly unit Heron.

The discount is the problem. For trigger conditions, "play" means the card resolved, and a countered spell never resolves (Rule 419.4.a.1). Since his Defy was countered, Astral Heron never triggered — so its [2][A][A] reduction doesn't exist and he pays Not So Fast's full cost. If Not So Fast then resolves, it becomes his first card played this turn, so Heron triggers and the discount gets banked for the next card he plays after that (FAQ #11762).