TL;DR: It's true. You have a good sense that positive semi-definiteness would help. We don't have it for the Paulis, but we have it for the eigenprojectors.
Let $P_+$ and $P_-$ denote the projectors onto the eigenspaces of $\mathcal{P}_i$ associated with the $+1$ and $-1$ eigenvalues, respectively, and similarly for $R_\pm$ and $\mathcal{P}_j$. Also, set $Q_\pm:=\mathcal{E}(R_\pm)$. Then
\begin{align}
\frac{1}{d}\mathrm{tr}(\mathcal{P}_i^\dagger\mathcal{E}(\mathcal{P}_j)&=\frac{1}{d}\mathrm{tr}((P_+-P_-)^\dagger\mathcal{E}(R_+-R_-))\tag1\\
&=\frac{1}{d}[\mathrm{tr}(P_+Q_+)-\mathrm{tr}(P_-Q_+)-\mathrm{tr}(P_+Q_-)+\mathrm{tr}(P_-Q_-)].\tag2
\end{align}
Now, $P_\pm$ and $R_\pm$ are positive semi-definite. Moreover, $\mathcal{E}$ is completely positive, so $Q_\pm$ are positive semi-definite, too. Hilbert-Schmidt inner product of two positive semi-definite operators is non-negative, so $\mathrm{tr}(P_\pm Q_\pm)\geqslant 0$. Therefore,
\begin{align}
-\mathrm{tr}(P_+Q_+)-\mathrm{tr}(P_-Q_+)-\mathrm{tr}(P_+Q_-)-\mathrm{tr}(P_-Q_-)\tag3\\
\leqslant\mathrm{tr}(P_+Q_+)-\mathrm{tr}(P_-Q_+)-\mathrm{tr}(P_+Q_-)+\mathrm{tr}(P_-Q_-)\tag4\\
\leqslant\mathrm{tr}(P_+Q_+)+\mathrm{tr}(P_-Q_+)+\mathrm{tr}(P_+Q_-)+\mathrm{tr}(P_-Q_-).\tag5
\end{align}
However,
\begin{align}
&\mathrm{tr}(P_+Q_+)+\mathrm{tr}(P_-Q_+)+\mathrm{tr}(P_+Q_-)+\mathrm{tr}(P_-Q_-)\tag6\\
&=\mathrm{tr}((P_++P_-)\mathcal{E}(R_++R_-))\tag7\\
&=\mathrm{tr}(\mathcal{E}(I))=2^n=d\tag8
\end{align}
where we used the fact that $\mathcal{E}$ is trace preserving. Substituting into the inequalities above and reintroducing normalization constant $\frac{1}{d}$, we obtain
\begin{equation}
-1\leqslant \frac{1}{d}\mathrm{tr}(\mathcal{P}_i^\dagger\mathcal{E}(\mathcal{P}_j))\leqslant+1\tag9
\end{equation}
as claimed.