v0.4.0 - multi-gpu and result-by-alias

What's Changed

• 🐛 Multi-GPU ensembles are possible again. Just extend the DeviceAllocatorContextManagerV2 context manager.
• ✨ Add return_alias_dict=True when calling predict_quantified and the return value will be a dict of quantifier aliases pointing to (prediction, confidence_or_uncertainty) tuples.