New DeepSeek-R1T-Chimera Model Merges R1 Reasoning with Efficiency of V3-0324