Bug in CreateTensorValueFromSystemNumericsTensorObject

The api CreateTensorValueFromSystemNumericsTensorObject throws the following error:

```
Unhandled exception. System.MissingMethodException: Method not found: 'Void System.Numerics.Tensors.Tensor`1.CopyTo(System.Numerics.Tensors.TensorSpan`1)'.
at Microsoft.ML.OnnxRuntime.OrtValue.CreateTensorValueFromSystemNumericsTensorObject[T](Tensor`1 tensor)
at Microsoft.ML.OnnxRuntime.OrtValue.CreateTensorValueFromSystemNumericsTensorObject[T](Tensor`1 tensor)
at Program.$(String[] args) in D:\Code\NET\ML\fai\Example\FailingTensor.cs:line 15
```

When running:
```C#
#:package System.Numerics.Tensors@10.0.0-preview.6.*
#:package Microsoft.ML.OnnxRuntime.Gpu@1.22.1

#:property NoWarn=$(NoWarn);SYSLIB5001

using System.Numerics.Tensors;
using Microsoft.ML.OnnxRuntime;

Span length = [1000, 1];
Span strides = [1, 0];

Tensor tensor = Tensor.Create(length, strides);

OrtValue.CreateTensorValueFromSystemNumericsTensorObject(tensor);
```

Also, looking at the implementation of CreateTensorValueFromSystemNumericsTensorObject, there is a bug/performance issue with `IsContiguousAndDense` -

If I have a tensor of shape [1000, 1] then the strides will be [1, 0] and contiguous and dense.
But then both checks fail, and it returns false even though it should return true.