I think that the basic problem here is the difference between normal modes and their frequencies.
Each normal mode corresponds to a well definite and independent displacement field of the atoms. There is the possibility of degenerate modes, i.e., independent modes with the same frequency. Therefore, the information that only three frequencies are present does not automatically imply only three normal modes.
However, before any conclusion from the analogy with the real CO$_2$ molecule can be drawn, we should know what exactly the hypotheses underlying the question are.
Here we have three different possible scenarios.
the motion is constrained in one dimension (the three atoms can only move along a straight line). This case is different from the case of a linear molecule in three dimensions (see next points). In this completely one-dimensional case, the system has only three independent vibrational modes. Remember that each normal mode corresponds to a linear combination of the original displacements. With three independent positions, we may have only three independent displacements and three independent modes. In such a case, there is no degenerate mode. The zero-frequency mode corresponds to the displacement such that no relative distance is changed (so no restoring force: all the atoms have the same displacement in the same direction). The two remaining modes (symmetric and asymmetric stretching) have different frequencies. Without explicit calculation of the eigenvectors of the dynamical matrix, the correct assignment of the two non-zero frequencies to the corresponding modes may be puzzling.
the motion is not constrained to one dimension (a full three-dimensional motion), and there are only two elastic nearest-neighbor interactions. In such a case, we have nine degrees of freedom, therefore, nine normal nodes. Being the molecule linear, five modes have zero frequency due to the global translational and rotational symmetry. Still, two additional zero-frequency modes exist, corresponding to the bending of the molecule around the central atom. The remaining two normal modes are like in the one-dimensional case.
finally, realistic models of the interactions introduce one additional elastic constant for angular deviations from the straight geometry, inducing two degenerate non-zero frequencies. These modes are absent in the model as shown in the figure but are present in a real diatomic molecule like CO$_2$.