feat: add Qwen Image 2.1 prefix KV cache

[leejet]

Summary

• Cache text and reference-image KV during the first denoising call, then compute only target-image tokens on subsequent calls.
• Keep separate FP32 caches for each condition within a sampling run, using existing RunnerCache lifecycle, graph cuts, and memory accounting.
• On allocation failure, release caches and retry once without caching. Bypass caching when generation extensions are
  enabled.

Related Issue / Discussion

Fix #2031.

Additional Information

.\bin\Release\sd-cli.exe --diffusion-model  ..\models\diffusion_models\qwen_image_2.1_int8_convrot.safetensors --vae ..\models\vae\qwen_image_2.1_vae_bf16.safetensors  --llm ..\models\text_encoders\Qwen3VL-8B-Instruct-Q4_K_M.gguf  -p "a lovely cat holding a sign says 'qwen2.1.cpp'" --cfg-scale 1.0 --sampling-method euler -v --offload-to-cpu --fa

8.34it/s -> 9.58it/s

.\bin\Release\sd-cli.exe --diffusion-model ..\models\diffusion_models\qwen_image_2.1_int8_convrot.safetensors --vae ..\models\vae\qwen_image_2.1_vae_bf16.safetensors --llm ..\models\text_encoders\Qwen3VL-8B-Instruct-Q4_K_M.gguf --llm_vision ..\models\text_encoders\Qwen3VL-8B-Instruct-mmproj-BF16.gguf -r ..\assets\qwen\qwen_image_2.1.png -p "change 'qwen2.1.cpp' to 'sd.cpp'" --cfg-scale 1.0 --sampling-method euler -v --offload-to-cpu

3.13it/s -> 5.84it/s

Checklist

• I have read and confirmed this PR follows the contribution guidelines.