feat(modelkit): add swin2sr support and canonical cpu recipe

[fangyangci]

Summary

• Add Swin2SR task/class registration and ONNX registration shim
• Wire Swin2SR mapping into HF aggregate registry
• Add canonical recipe at examples/recipes/caidas_swin2SR-classical-sr-x2-64/cpu/cpu/image-to-image_fp16_config.json
• Add unit tests for swin2sr support and loader task coverage

Validation

• uv run pytest tests/unit/models/test_swin2sr_support.py tests/unit/loader/test_get_supported_tasks.py -q
• uv run python -c "import json; from pathlib import Path; from winml.modelkit.config import WinMLBuildConfig; p=Path('examples/recipes/caidas_swin2SR-classical-sr-x2-64/cpu/cpu/image-to-image_fp16_config.json'); d=json.loads(p.read_text(encoding='utf-8')); WinMLBuildConfig.from_dict(d); print('recipe-parse-ok')"
• uv run winml inspect --model-type swin2sr --list-tasks