Exploring Embedding Strategies for Heterogeneous Graph Representation Learning with GNNs