Torch var not found in context - although variable is class attribute

## ❓Converting simple PyTorch model via TorchScript to CoreML fails, as a member variable cannot be found in forward() call

## System Information
- Mac Mini M1 (2020)
- MacOS 11.6
- Python 3.9
- PyTorch 1.9.1
- CoreMLTools 5.1.0

I am a newbie in converting models and maybe someone please can tell me, why the *self.attribute1* value can not be found in the *forward()* call? How can I work around that error?

```python
import torch
import torch.nn as nn
from torch import Tensor
from typing import Optional, Dict, List
import coremltools as ct

print(torch.__version__) # 1.9.1
print(ct.__version__) # 5.1.0

class SimpleTest(nn.Module):
def __init__(self) -> None:
super().__init__()
self.attribute1: float = 42.

def forward(self, x: Tensor) -> Tensor:
fillval: float = 42. # this alone works
fillval: float = self.attribute1 # ValueError: Torch var fillval.3 not found in context
some_result = torch.full((3, 550, 550), fillval)
return some_result

test_batch = torch.rand(1,3,550,550)
m = SimpleTest()
m.eval()
result = m(test_batch)
print('1 Before torch-script -------------------')
s = torch.jit.script(m)
print('2 Before coreml-convert -------------------')
c = ct.convert(s,
inputs=[ct.TensorType(name="test", shape=test_batch.shape)],
source='auto',
minimum_deployment_target=ct.target.iOS15,
compute_units=ct.ComputeUnit.CPU_ONLY,
compute_precision=ct.precision.FLOAT32,
convert_to='mlprogram',
debug=True
)

```

Run the provided Python example with the listed PyTorch 1.9.1 and CoreMLTools 5.1.0 versions, starting at SimpleTest.forward, torch.jit.script(m), and ct.convert(s). Compare the literal fillval case with the self.attribute1 case; done means the reported context error is explained and the conversion behavior or workaround is verified.

Written by the indexing model from the issue text.