ct.convert call appears to corrupt torchscript model

## 🐞Describing the bug
After running `ct.convert` on a torchscript model, the torchscript model appears to be corrupted and does not save correctly. The stack trace is coming from torch, but it only happens after the model has been processed using ct.convert.

## Stack Trace
```
python version: 3.9.19 (main, May 6 2024, 14:39:30)
[Clang 14.0.6 ]
torch version: 2.2.0
ct version: 7.2
** model loaded correctly before ct.convert
Converting PyTorch Frontend ==> MIL Ops: 75%|███████████████████████████████████████████████████▊ | 3/4 [00:00<00:00, 1289.10 ops/s]
Running MIL frontend_pytorch pipeline: 100%|████████████████████████████████████████████████████████████████████| 5/5 [00:00<00:00, 6848.96 passes/s]
Running MIL default pipeline: 0%| | 0/78 [00:00
_ = torch.jit.load(f1)
File "/Users/carson/miniconda3/envs/ct_convert_error/lib/python3.9/site-packages/torch/jit/_serialization.py", line 159, in load
cpp_module = torch._C.import_ir_module(cu, str(f), map_location, _extra_files, _restore_shapes) # type: ignore[call-arg]
RuntimeError: required keyword attribute 'chunks' is undefined
```

## To Reproduce
Python script:
```
import torch
import torch.nn as nn
import coremltools as ct
import numpy as np
import sys

f0 = 'tmp0.pt'
f1 = 'tmp1.pt'

print(f'python version: {sys.version}')
print(f'torch version: {torch.__version__}')
print(f'ct version: {ct.__version__}')

class Net(nn.Module):
def __init__(self):
super().__init__()

def forward(self, x):
a,b,c = x.chunk(3)
return (a * b) + c

with torch.no_grad():

i = torch.rand((768, 256))
net = Net().eval()
net_traced = torch.jit.trace(net, (i))

# this works..
net_traced.save(f0)
_ = torch.jit.load(f0)
print('** model loaded correctly before ct.convert')

ct.convert(
net_traced,
convert_to='mlprogram',
minimum_deployment_target=ct.target.macOS12,
compute_units=ct.ComputeUnit.ALL,
inputs=[ct.TensorType(name='i0', shape=(768,25), dtype=np.float32)])

# this doesnt..
net_traced.save(f1)
_ = torch.jit.load(f1)
print('** model loaded correctly after ct.convert')
```

## System environment (please complete the following information):
- coremltools version: 7.2
- OS (e.g. MacOS version or Linux type): macOS 14, M1
- Any other relevant version information (e.g. PyTorch or TensorFlow version): pytorch: 2.2.0

Start with the reproduction in repro.py and run the stated PyTorch 2.2.0/coremltools 7.2 example, comparing torch.jit.load(f0) and torch.jit.load(f1) around ct.convert. Trace the PyTorch conversion entry point and the warning location in coremltools/converters/mil/mil/passes/defs/preprocess.py; done means saving after conversion no longer prevents torch.jit.load from succeeding.

Written by the indexing model from the issue text.