Rewrite README: document all variants (f16/bf16/Q4_K_M/NF4 x cuda/default/onnx-standard/webgpu + OpenVINO)

README: EPContext verified end-to-end; add runtime wheels (OV 2026.3 + ORT 1.28 OpenVINO EP)

README: document openvino-epcontext variant

Add OpenVINO EPContext models (ORT-loadable compiled OV graph) for gemma-4-E2B-it

Delete openvino

Update README: add OpenVINO IR variant (OV 2026.3), path now works

Add OpenVINO IR (compiled artifact) for gemma-4-E2B-it, built with OpenVINO 2026.3 (incl. RMSNorm/RoPE opset-24 translators)

Add README (OpenVINO INT4 usage + opset-24 Attention limitation)

Add OpenVINO NPU INT4 model (mobius + K-Quant) for google/gemma-4-E2B-it

Rebuild NF4/cuda (quantized decoder) with per_layer_inputs contract

Rebuild NF4/default (quantized decoder) with per_layer_inputs contract

Rebuild Q4_K_M/cuda (quantized decoder) with per_layer_inputs contract

Rebuild Q4_K_M/default (quantized decoder) with per_layer_inputs contract

Rebuild bf16/onnx-standard with per_layer_inputs contract

Rebuild bf16/cuda with per_layer_inputs contract

Rebuild bf16/default with per_layer_inputs contract

Rebuild f16/onnx-standard with per_layer_inputs contract

Rebuild f16/cuda with per_layer_inputs contract

Rebuild f16/default with current per_layer_inputs contract

Add f16 webgpu build (gemma-4-E2B-it)

Add NF4/cuda CUDA EP quantized variant

Add Q4_K_M/cuda CUDA EP quantized variant

Remove safetensors files (replaced by onnx external data)

Rebuild NF4/default with onnx external data (latest mobius)

Rebuild Q4_K_M/default with onnx external data (latest mobius)

Rebuild bf16/onnx-standard with onnx external data (latest mobius)

Rebuild bf16/cuda with onnx external data (latest mobius)

Rebuild bf16/default with onnx external data (latest mobius)

Rebuild f16/onnx-standard with onnx external data (latest mobius)

Rebuild f16/cuda with onnx external data (latest mobius)

Rebuild f16/default with onnx external data (latest mobius)

Re-export NF4/default with onnx external data

Re-export Q4_K_M/default with onnx external data

Re-export bf16/onnx-standard with onnx external data format

Re-export bf16/cuda with onnx external data format

Re-export bf16/default with onnx external data format

Re-export f16/onnx-standard with onnx external data format

Re-export f16/cuda with onnx external data format

Re-export f16/default with onnx external data format

Remove old int4-nf4 (renamed)

Remove old int4-kquant (renamed)

Add Q4_K_M quantized variant

Add NF4 quantized variant

Add int4-kquant quantized variant

Add int4-kquant quantized variant

Add int4-nf4 quantized variant

Remove old flat file chat_template.jinja

Remove old flat file audio_feature_extraction.json

Remove old flat file tokenizer_config.json

Remove old flat file tokenizer.json