Can PGM index and linear approximation models in general be applied to clustered indexes, where actual data of variable size are stored in the index along with the keys?
Yep, in that case you could use an indirection vector containing, for each key k, the offset to the first byte of k.
This is what is typically done in B-trees, where the indirection vector is stored in the header of a disk page.
It's described for example in Section 3.3 "Variable-length records" of Goetz Graefe's "Modern B-Tree Techniques".