MIL program object is not persisted between saving and loading a MLModel object

```python
import coremltools as ct
import numpy as np
import torch

# Create simple torch model and convert to mlprogram
x = torch.rand(10)
m_torch = torch.jit.trace(torch.nn.Linear(10, 2), x)
m1 = ct.convert(m_torch, inputs=[ct.TensorType(shape=x.shape)], convert_to="mlprogram")

# Model is valid; can get predictions.
print(m1.predict({'input': x}))

# MIL is present.
print(m1._get_mil_internal())

# Save Core ML model to disk and read it.
path = '/tmp/test.mlpackage'
m1.save(path)
m2 = ct.models.MLModel(path)

# MIL is not present!
print(m2._get_mil_internal())
```

Start by reproducing the provided Python example, focusing on MLModel.save, ct.models.MLModel(path), and _get_mil_internal(). Trace how the MIL program is handled across save and load, then add a regression test showing that the loaded model retains the MIL program.

Written by the indexing model from the issue text.