Ok, the issue is that I'm in fact very stupid: the covariant Yoneda embedding takes values in $\mathbf{Set}^{\mathcal{C}^{op}}$. It is then clear that a functor $F\in\mathbf{Set}^{\mathcal{C}^{op}}$ is the *co*limit of $\operatorname{el}(F)\rightarrow\mathcal{C}\rightarrow\mathbf{Set}^{\mathcal{C}^{op}}$. This gives us the Yoneda embedding as free cocompletion.
Then we dualize and get that $(\mathbf{Set}^{(\mathcal{C}^{op})^{op}})^{op}=(\mathbf{Set}^{\mathcal{C}})^{op}$ is the opposite of the free cocompletion of $\mathcal{C}^{op}$, so the free completion of $\mathcal{C}$.