[Feature Request] QLinearMatmul : weight zero point must be a scalar, 1D tensor of size 1, or last to second dimension is 1

### Describe the feature request

onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException:
[ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running QLinearMatMul node. Name:'MatMul_266' Status Message: /onnxruntime_src/onnxruntime/core/providers/cpu/math/quantize_linear_matmul.cc:37 virtual onnxruntime::common::Status onnxruntime::QLinearMatMul::Compute(onnxruntime::OpKernelContext*) const IsBQuantParamSupported(b_offset->Shape(), b ? b->Shape() : b_shape_) was false. QLinearMatmul : weight zero point must be a scalar, 1D tensor of size 1, or last to second dimension is 1

### Describe scenario use case


Thanks in advance!

Start with onnxruntime/core/providers/cpu/math/quantize_linear_matmul.cc and inspect QLinearMatMul::Compute, especially the IsBQuantParamSupported check named in the error. Determine how weight zero-point shapes are validated, then confirm the requested scalar, one-element 1D, and supported last-to-second-dimension forms are accepted without regressing existing behavior.

Written by the indexing model from the issue text.