MatryoshkaLoss

This release focuses on correctness and performance, rebuilding gradient-cached losses on a shared engine for faster training and fixing several gradient bugs. It also significantly improves inference speed when using `model.compile()`.

Sentence‑Transformers 5.1.2 adds improved saving for StaticEmbedding and Dense modules, introduces Intel XPU as the default device, enhances loss compatibility, and adds Python 3.13 support while fixing several training and loading bugs.