Updated model for ORT support

[varun-jaiswal17]

Summary

This PR patches 18 ONNX test models under testdata/dnn/onnx/models/ to be
compatible with the ONNX Runtime (ORT) engine (OPENCV_FORCE_DNN_ENGINE=4).
The models were regenerated or attribute-patched to remove deprecated/invalid
ONNX attributes that ORT enforces strictly.

Companion PR: opencv/opencv#28872

---

Binary model changes

Model file	Change
average_pooling.onnx	Removed deprecated kernel_shape / strides attribute encoding; re-exported at opset 7
batch_norm.onnx	Removed legacy spatial, is_test, consumed_inputs attributes (opset 1 → 7)
constant.onnx	Cleaned up unused legacy attributes; re-exported at opset 7
dropout.onnx	Removed is_test, consumed_inputs attributes (opset 1 → 7)
linear.onnx	Removed legacy broadcast attribute from Add node (opset 1 → 7)
mul.onnx	Removed legacy broadcast, axis attributes from Mul node (opset 1 → 7)
multy_inputs.onnx	Re-exported to remove deprecated node attributes
equal_same_dims.onnx	Re-exported to clean up opset metadata
reduce_mean_axis1.onnx	Replaced deprecated axes attribute with axes input (opset 11 → 11, attribute-style fixed)
reduce_mean_axis2.onnx	Replaced deprecated axes attribute with axes input (opset 11 → 11, attribute-style fixed)
reduce_sum.onnx	Moved axes to input tensor per opset 13 semantics
dynamic_resize_10.onnx	Added explicit sizes / scales input to satisfy ORT shape requirements
dynamic_resize_9.onnx	Added required inputs for Resize opset 9
dynamic_resize_scale_10.onnx	Added explicit scales input
dynamic_resize_scale_9.onnx	Added explicit scales input
resize_bilinear.onnx	Added required roi and scales inputs (opset 11)
resize_nearest.onnx	Added required scales input (opset 9)
resize_nearest_unfused_opset11_torch1.3.onnx	Simplified model; removed extraneous subgraph nodes that ORT rejected

---

Testing

All 18 affected tests pass with OPENCV_FORCE_DNN_ENGINE=4 , OPENCV_FORCE_DNN_ENGINE=2 , OPENCV_FORCE_DNN_ENGINE=1:

AveragePooling, BatchNormalization, Dropout, Linear, Multiplication,
CompareSameDims_EQ, ReduceMean, ReduceSum, Quantized_MatMul (via constant.onnx),
dynamic_resize_, resize_bilinear, resize_nearest, resize_nearest_unfused_