data: kvKbPerToken column (exact KV-cache cost for hybrid linear-attention models)

data: 107 models - adds Xiaomi MiMo-V2-Flash + ggufDiy column

docs: document openWeights field + runsLocally semantics

Create README.md

initial commit