Fix quant unpatch (#1129)

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

- **New Features**
- Introduced unpatching functionality for quantized models, enhancing
model management.
- Added quantization tracking to the OneflowDeployableModule, improving
configuration capabilities.
	- Implemented a method to check the quantization state of the module.

- **Bug Fixes**
- Improved handling of unpatching conditions for quantized models to
prevent unintended skips.

- **Documentation**
- Updated documentation for the apply_online_quant method with usage
examples.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>

onediff_comfy_nodes/modules/oneflow/hijack_model_patcher.py

@@ -1,4 +1,5 @@
 from comfy.model_patcher import ModelPatcher
+from onediff.utils import logger
 
 from ..sd_hijack_utils import Hijacker
 from .patch_management import create_patch_executor, PatchType
@@ -18,6 +19,25 @@ def cond_func(org_fn, self):
     return is_using_oneflow_backend(self)
 
 
+def unpatch_model_oneflow(org_fn, self, device_to=None, unpatch_weights=True):
+    if unpatch_weights:
+        logger.warning(
+            f"{type(self.model.diffusion_model)} is quantized by onediff, so unpatching is skipped."
+        )
+    return
+
+
+def unpatch_model_cond_func(org_fn, self, *args, **kwargs):
+    if hasattr(self.model, "diffusion_model") and hasattr(
+        self.model.diffusion_model, "_deployable_module_quantized"
+    ):
+        return self.model.diffusion_model._deployable_module_quantized
+    return False
+
+
 model_patch_hijacker = Hijacker()
 
 model_patch_hijacker.register(ModelPatcher.clone, clone_oneflow, cond_func)
+model_patch_hijacker.register(
+    ModelPatcher.unpatch_model, unpatch_model_oneflow, unpatch_model_cond_func
+)

src/onediff/infer_compiler/backends/oneflow/deployable_module.py

@@ -247,6 +247,7 @@ def apply_online_quant(self, quant_config):
             >>> model.apply_online_quant(quant_config)
         """
         self._deployable_module_quant_config = quant_config
+        self._deployable_module_quantized = True
 
 
 def get_mixed_deployable_module(module_cls):