Align patch_size between image and video processing

Image processor uses `patch_size=2` but video processor has `patch_size=16`. This misalignment leads to error (`RuntimeError: mat1 and mat2 shapes cannot be multiplied (20160x768 and 12x8)`) when running the model with transformers and video input.

processor_config.json

@@ -69,7 +69,7 @@
     ],
     "max_soft_tokens": 70,
     "num_frames": 32,
-    "patch_size": 16,
+    "patch_size": 2,
     "pooling_kernel_size": 3,
     "resample": 3,
     "rescale_factor": 0.00392156862745098,