Add measured Galaxy S26 NPU/GPU section

Card: add a measured Performance table for the representative graphs (Pixel 8a, benchmark_model GPU+CPU)

Document typed prompts (tokenizer + embedding table)

Add prompt-encoding asset qwen_embed_fp16.bin

Add prompt-encoding asset tokenizer_fixture.txt

Add prompt-encoding asset qwen_special.txt

Add prompt-encoding asset qwen_merges.txt

Add prompt-encoding asset qwen_vocab.txt

Add the editing hero image

Document image editing (kce_* graphs + VAE encoder)

Add image-editing graph kv_vae_enc.tflite

Add image-editing graph kce_final.tflite

Add image-editing graph kce_single3.tflite

Add image-editing graph kce_single2.tflite

Add image-editing graph kce_single1.tflite

Add image-editing graph kce_single0.tflite

Add image-editing graph kce_double1.tflite

Add image-editing graph kce_double0.tflite

Add image-editing graph kce_prep.tflite

Fix GPU name: Pixel 8a is Mali-G715 (Tensor G3), not Mali-G610

Add FLUX.2-klein-4B int8 LiteRT graphs (12 chunks, GPU-clean)

initial commit