torch.where behaves strangely with CoreML via ExecuTorch

When a model has torch.where(x, min, max), the min and max needs to be registered within the model for it to work. Otherwise, the model output is wrong on iOS. When inferencing on iOS with CoreML via ExecuTorch, model produces non-meaningful results compared to python inference.

This won't work:
```
class Model(nn.Module):
def __init__(self, ):
super(Model, self).__init__()
self.conv = nn.Conv2d(3, 3, 3, 1, 1, bias=True)

def forward(self, x):
x = self.conv(x)
x = torch.where(x > 0.5, 0.0, 1.0)
return x

```

This will work:

```
class Model(nn.Module):
def __init__(self, ):
super(Model, self).__init__()
self.conv = nn.Conv2d(3, 3, 3, 1, 1, bias=True)

def forward(self, x):
x = self.conv(x)
x = torch.where(x > 0.5, self.lower, self.upper)
return x

model = Model()
upper = torch.tensor(1.0).to(float_dtype)
lower = torch.tensor(0.0).to(float_dtype)
model.register_buffer("upper", upper)
model.register_buffer("lower", lower)
```

Start with the provided torch.where reproduction and compare its Python output with the iOS CoreML output through ExecuTorch. Trace how scalar min and max operands are represented during conversion, and consider the issue done when the reproduction produces matching, meaningful outputs without requiring registered buffers.

Written by the indexing model from the issue text.