Remove stale v2 npu_fp16 engines (R2 uses npu_int4)

Deploy R2: TRT/OV/TVM engines + ONNX(fp32+int8) + SP tokenizer

Model card: R2 (gte-modernbert distill) + reranker section

card: neutral what-is-in-repo wording

card: model.onnx + INT8 ONNX are bundled (fix not-bundled note)

Add ONNX: FP32 model.onnx (+data) + INT8 model_int8qdt.onnx (engine-build source)

card: accurate file list + tokenizer (SP) note

Add compiled engines: TRT win sm_120 + OV (cpu/igpu/npu) + TVM win vulkan + tokenizer

Upload README.md with huggingface_hub

initial commit