The challenges associated with channel state information (CSI) acquisition in the frequency division duplexing (FDD) extremely large-scale MIMO (XL-MIMO) system significantly impede its application in 6G communications. In such context, this contribution introduces a CSI feedback framework with pilot precoding capitalizing on the partial FDD reciprocity toward significantly enhanced efficiency. Specifically, by exploiting the near-field channel sparsity in the polar-delay domain, the proposed polar-delay sparsity (PDS) codebook remarkably reduces pilot and CSI feedback overhead while maintaining high spectral efficiency (SE). Furthermore, a class of innovative compression and decompression methods are also developed to enable further reduction of training overhead without compromising system performance. Notably, the proposed design ensures consistently low computational complexity and feedback overhead at the user equipment (UE), making it well-suited for the massive connectivity demands of heterogeneous devices in 6G Internet of Everything (IoE) applications. Finally, simulation results demonstrate the satisfactory SE performance enhancement achieved by the proposed schemes.