Yes. When increasing the vertex buffer, you also need to modify the index buffer.

It would make matters easier when your texture mapping would map all parts of the model without overlaps. This way the light mapping does not need extra duplicated vertices. The lightmaps I've seen so far contained less vertices than the texture map, and also did not contain the gaps and separated parts of your above example.

In such a case it would probably be easier to create the mesh from the lightmap, and import the texture uv afterwards - not vice versa.