Fix save_pretrained / from_pretrained round-trip for KimiK25VisionProcessor

## Problem

``processor.save_pretrained()`` serializes transient Hub kwargs (``revision``, ``_from_auto``, ``cache_dir``, ``force_download``, ``local_files_only``) into ``preprocessor_config.json``. On reload, ``from_pretrained`` passes these same kwargs again, and ``from_dict`` merges ``**config`` with ``**kwargs`` — causing a ``TypeError: got multiple values for keyword argument``:

```python
processor = AutoProcessor.from_pretrained("moonshotai/Kimi-K2.6")
processor.save_pretrained("/tmp/test")
AutoProcessor.from_pretrained("/tmp/test", trust_remote_code=True)
# TypeError: KimiK25VisionProcessor() got multiple values for keyword argument 'revision'
```

## Fix

Strip transient loading kwargs in ``__init__``, ``from_dict``, and ``to_dict`` so they are never stored as instance attributes or serialized.

kimi_k25_vision_processing.py

@@ -22,6 +22,21 @@ try:
 except ImportError:
     VideoReader = None
 
+_TRANSIENT_KWARGS = frozenset({
+    "_from_auto",
+    "cache_dir",
+    "force_download",
+    "local_files_only",
+    "revision",
+    "token",
+    "proxies",
+    "resume_download",
+})
+
+
+def _pop_transient(d: dict) -> dict:
+    return {k: v for k, v in d.items() if k not in _TRANSIENT_KWARGS}
+
 
 def resampling(video_bytes: bytes,
                sample_indices: list[int],
@@ -46,6 +61,7 @@ class KimiK25VisionProcessor(BaseImageProcessor):
         media_proc_cfg: dict,
         **kwargs,
     ):
+        kwargs = _pop_transient(kwargs)
         super().__init__(**kwargs)
         self.media_proc_cfg = media_proc_cfg
         self.num_frames_per_chunk = media_proc_cfg[
@@ -162,11 +178,11 @@ class KimiK25VisionProcessor(BaseImageProcessor):
     ) -> BatchFeature:
         """
         Preprocess a atom vision input (images/video_chunk) into model-ready tensors.
-        
+
         Args:
             medias: List of MediaInput.
             return_tensors: Desired output format ('pt', 'np', 'tf', or None).
-        
+
         Returns:
             BatchFeature containing 'pixel_values' and 'grid_thws' tensors.
         """
@@ -232,6 +248,7 @@ class KimiK25VisionProcessor(BaseImageProcessor):
 
     def to_dict(self) -> Dict[str, Any]:
         output = super().to_dict()
+        output = _pop_transient(output)
         output["media_proc_cfg"] = self.media_proc_cfg
         if "media_processor" in output:
             del output["media_processor"]
@@ -239,7 +256,8 @@ class KimiK25VisionProcessor(BaseImageProcessor):
 
     @classmethod
     def from_dict(cls, config_dict: Dict[str, Any], **kwargs):
-        config = config_dict.copy()
+        config = _pop_transient(config_dict.copy())
+        kwargs = _pop_transient(kwargs)
         media_proc_cfg = config.pop("media_proc_cfg", {})
         return cls(media_proc_cfg=media_proc_cfg, **config, **kwargs)