Pre-initialize torch._dynamo to prevent double-registration with peft torch.compile() call (#1228)

### What does this PR do?

Type of change: ? Bug fix

This PR fixes the error `AssertionError: Artifact of type=precompile
already registered in mega-cache artifact factory` when launching with
local docker.

The error message:
```
Qwen3-8B-0/0 Traceback (most recent call last):
Qwen3-8B-0/0   File "/nemo_run/code/modules/Megatron-LM/examples/post_training/modelopt/quantize.py", line 21, in <module>
Qwen3-8B-0/0     import modelopt.torch.quantization as mtq
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/modelopt/torch/__init__.py", line 23, in <module>
Qwen3-8B-0/0     from . import distill, nas, opt, peft, prune, quantization, sparsity, speculative, utils
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/modelopt/torch/prune/__init__.py", line 24, in <module>
Qwen3-8B-0/0     from . import fastnas, gradnas, plugins
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/modelopt/torch/prune/gradnas.py", line 71, in <module>
Qwen3-8B-0/0     from transformers.models.bert.modeling_bert import BertAttention
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/transformers/models/bert/modeling_bert.py", line 30, in <module>
Qwen3-8B-0/0     from ...generation import GenerationMixin
Qwen3-8B-0/0   File "<frozen importlib._bootstrap>", line 1412, in _handle_fromlist
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/transformers/utils/import_utils.py", line 2317, in __getattr__
Qwen3-8B-0/0     module = self._get_module(self._class_to_module[name])
Qwen3-8B-0/0              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/transformers/utils/import_utils.py", line 2347, in _get_module
Qwen3-8B-0/0     raise e
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/transformers/utils/import_utils.py", line 2345, in _get_module
Qwen3-8B-0/0     return importlib.import_module("." + module_name, self.__name__)
Qwen3-8B-0/0            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Qwen3-8B-0/0   File "/usr/lib/python3.12/importlib/__init__.py", line 90, in import_module
Qwen3-8B-0/0     return _bootstrap._gcd_import(name[level:], package, level)
Qwen3-8B-0/0            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/transformers/generation/utils.py", line 43, in <module>
Qwen3-8B-0/0     from ..masking_utils import create_masks_for_generate
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/transformers/masking_utils.py", line 40, in <module>
Qwen3-8B-0/0     from torch._dynamo._trace_wrapped_higher_order_op import TransformGetItemToIndex
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/torch/_dynamo/__init__.py", line 13, in <module>
Qwen3-8B-0/0     from . import (
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/torch/_dynamo/aot_compile.py", line 16, in <module>
Qwen3-8B-0/0     from torch._dynamo.package import SystemInfo
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/torch/_dynamo/package.py", line 443, in <module>
Qwen3-8B-0/0     @CacheArtifactFactory.register
Qwen3-8B-0/0      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Qwen3-8B-0/0   File "/usr/local/lib/python3.12/dist-packages/torch/compiler/_cache.py", line 72, in register
Qwen3-8B-0/0     assert artifact_cls.type() not in cls._artifact_types, (
Qwen3-8B-0/0            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Qwen3-8B-0/0 AssertionError: Artifact of type=precompile already registered in mega-cache artifact factory
```

### Before your PR is "*Ready for review*"

Make sure you read and follow [Contributor
guidelines](https://github.com/NVIDIA/Model-Optimizer/blob/main/CONTRIBUTING.md)
and your commits are signed (`git commit -s -S`).

Make sure you read and follow the [Security Best
Practices](https://github.com/NVIDIA/Model-Optimizer/blob/main/SECURITY.md#security-coding-practices-for-contributors)
(e.g. avoiding hardcoded `trust_remote_code=True`, `torch.load(...,
weights_only=False)`, `pickle`, etc.).

- Is this change backward compatible?: ✅ 
- If you copied code from any other sources or added a new PIP
dependency, did you follow guidance in `CONTRIBUTING.md`: ❌
- Did you write any new necessary tests?: ❌
- Did you update
[Changelog](https://github.com/NVIDIA/Model-Optimizer/blob/main/CHANGELOG.rst)?:
❌




<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **Bug Fixes**
* Improved startup compatibility with certain Torch-based optimization
tools by ensuring required runtime components are loaded during package
initialization to prevent initialization errors.
* No changes to public APIs or exported modules; behavior and interfaces
remain unchanged.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: Hung-Yueh Chiang <hungyuehc@nvidia.com>

Author: hychiang-git

modelopt/torch/__init__.py

@@ -15,12 +15,25 @@
 
 """Model optimization and deployment subpackage for torch."""
 
+import importlib
 import warnings as _warnings
 
 from packaging.version import Version as _Version
 from torch import __version__ as _torch_version
 
-from . import distill, nas, opt, peft, prune, quantization, sparsity, speculative, utils
+# Pre-initialize torch._dynamo to prevent double-registration with peft's torch.compile() call
+importlib.import_module("torch._dynamo")
+from . import (  # noqa: E402
+    distill,
+    nas,
+    opt,
+    peft,
+    prune,
+    quantization,
+    sparsity,
+    speculative,
+    utils,
+)
 
 if _Version(_torch_version) < _Version("2.9"):
     _warnings.warn(