How to Move Your Dragon: Text-to-Motion Synthesis for Large-Vocabulary Objects