Every FLOP Counts: Scaling a 300B Mixture-of-Experts LING LLM without Premium GPUs