[AArch64] Use indexed dup for 128b segmented splat (#144688) Matches a splat of 128b segments into a wider z register expressed as a concat_vectors sdnode and generate a dup zn.q, zd.q[0] instruction. GitOrigin-RevId: 152d4b8a01e8671a676e7cfaf71c70b1edeee7e8