server, webui: support continue generation on reasoning models by ServeurpersoCom · Pull Request #22727 · ggml-org/llama.cpp

Finally, efficient parallel bulk generation with large input data (especially when paired with -kvu). If the context limit hits - just store the temporary result, retry later when more is free, instead of throwing it all away.