Remove deprecated `"qdq"` mode value from config and recipes

[DingmaomaoBJTU]

Summary

The WinMLQuantizationConfig.mode field currently accepts "qdq" as a legacy value, which is silently mapped to "static" in from_dict(). This creates confusion since "qdq" and "static" mean the same thing.

A follow-up PR should:

1. Update all recipe JSON files (~50+ in examples/recipes/) from "mode": "qdq" → "mode": "static"
2. Update documentation (docs/reference/index.md, docs/concepts/config-and-build.md) to remove "qdq" from the schema
3. Update unit tests (test_config.py, test_build.py, test_onnx.py, test_hf.py) to use "static" instead of "qdq"
4. Remove the backward-compat shim in src/winml/modelkit/quant/config.py (if raw_mode == "qdq": raw_mode = "static")
5. Update the Literal type from Literal["static", "dynamic", "rtn", "fp16"] — confirm no references to "qdq" remain

Notes

• The "qdq" value was the original default before the mode field was unified in feat: precision-driven quantization (FP16, RTN int4, static QDQ) via --precision flag #872
• Keep the shim in place until all recipes/docs/tests are migrated in the same PR to avoid a breaking window
• Consider whether winml config generation still emits "qdq" anywhere