Be careful claiming it is wait-free

You claim your queue is wait-free, but you have a potentially unbounded for-loop doing a compare-exchange operation. At the very least, this means consumer_get_head()'s time complexity is actually \$O(N)\$. Sure, you are unlikely to hit that case, but the whole point of a wait-free algorithm is that you never have to wait.

Confusing API

The API is very confusing. As you mention in the comments, the producer already has to know the pointer to the first free element, and then producer_force_put() will mark that element as produced and returns a pointer to the next free element. And something similar is going on with the consumer functions. For me, that was very unexpected. It also means the caller now shares some of the bookkeeping.

I would rather have a function producer_get_head() that returns a pointer to the first free element, and a producer_force_get_head() to force it to free up an element if there is none, and then a producer_put() that returns void and that merely marks the element as produced and increments the head index.