Update on "Add preprocess to ci"

1. Run `test_preprocess.py` in ci.
2. Refactor `test_preprocess.py` by creating the model artifacts (eager, exported, et) once and re-use them on different inputs (previously, create model artifacts for each test).

Test Plan:
```
python -m unittest examples.models.llama3_2_vision.preprocess.test_preprocess
```

[ghstack-poisoned]

Author: lucylq

examples/models/llama3_2_vision/preprocess/model.py

@@ -26,6 +26,9 @@ class PreprocessConfig:
     max_num_tiles: int = 4
     tile_size: int = 224
     antialias: bool = False
+    # Used for reference eager model from torchtune.
+    resize_to_max_canvas: bool = False
+    possible_resolutions: Optional[List[Tuple[int, int]]] = None
 
 
 class CLIPImageTransformModel(EagerModelBase):

examples/models/llama3_2_vision/preprocess/test_preprocess.py

@@ -59,11 +59,11 @@ def initialize_models(resize_to_max_canvas: bool) -> Dict[str, Any]:
         reference_model = CLIPImageTransform(
             image_mean=config.image_mean,
             image_std=config.image_std,
-            resize_to_max_canvas=config.resize_to_max_canvas,
             resample=config.resample,
             antialias=config.antialias,
             tile_size=config.tile_size,
             max_num_tiles=config.max_num_tiles,
+            resize_to_max_canvas=config.resize_to_max_canvas,
             possible_resolutions=None,
         )