Mitigating Capability Degradation in LLM Self-Improvement via Representation Geometry Regularization | AIChainDay