fix: re-export 20m weights; true int8/int4 keys; honest model card

docs: update Model Card with full sister models family and quantization guide

docs: add Hugging Face YAML metadata with jev, hastejev, system-1, and agent tags

feat: update Haste Jev weights, SimHash subword projection, and browser control benchmarks

feat: release Haste Jev weights (safetensors, bin), config.json, and model package

initial commit