Miles v0.1.2 adds experimental Kubernetes support and a third training backend

The Miles v0.1.2 announcement says its experimental Kubernetes backend lets reinforcement-learning jobs run as cluster workloads while orchestration restarts without stopping training. It says score centering keeps asynchronous training stable and GPUs busy, and PyTorch's torchtitan joins Megatron and FSDP as a third training backend. The announcement also lists DeepSeek-V4.1-Flash and MiMo-V2.6-Flash-RL as landing in main.