fix: transformers 5.x compat (cache_position + kwargs naming)

## Summary

This PR fixes two issues that prevent dots.ocr from working with transformers>=5.0:

### 1. cache_position TypeError on generation

In transformers 5.x, cache_position is no longer maintained in the generation loop. The current code does `cache_position[0] == 0` which crashes with `TypeError: 'NoneType' object is not subscriptable`.

Fix: Use a combined check compatible with both transformers 4.x and 5.x — fall back to `past_key_values is None` when cache_position is unavailable.

### 2. _validate_model_kwargs ValueError for processor outputs

forward() uses `**loss_kwargs` instead of `**kwargs`. Transformers 5.x validation only recognizes `**kwargs`/`**model_kwargs` as catch-all params, causing processor outputs like mm_token_type_ids to fail validation.

Fix: Rename `**loss_kwargs` to `**kwargs` (functionally identical).

### Backward compatibility

Both fixes maintain full backward compatibility with transformers 4.x.

modeling_dots_ocr.py

@@ -80,7 +80,7 @@ class DotsOCRForCausalLM(Qwen2ForCausalLM):
         return_dict: Optional[bool] = None,
         use_cache: Optional[bool] = None,
         logits_to_keep: int = 0,
-        **loss_kwargs,
+        **kwargs,
     ) -> Union[Tuple, CausalLMOutputWithPast]:
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
         assert len(input_ids) >= 1, f"empty input_ids {input_ids.shape=} will cause gradnorm nan"
@@ -99,7 +99,7 @@ class DotsOCRForCausalLM(Qwen2ForCausalLM):
             output_hidden_states=output_hidden_states,
             # return_dict=return_dict,
             logits_to_keep=logits_to_keep,
-            **loss_kwargs,
+            **kwargs,
         )
 
         return outputs
@@ -125,7 +125,15 @@ class DotsOCRForCausalLM(Qwen2ForCausalLM):
             **kwargs,
         )
 
-        if cache_position[0] == 0:
+        # Pass pixel_values only on the first generation step (prefill).
+        # Compatible with both transformers 4.x (cache_position available)
+        # and 5.x (cache_position removed, use past_key_values instead).
+        is_prefill = (
+            (cache_position is not None and cache_position[0] == 0)
+            or past_key_values is None
+            or (hasattr(past_key_values, "get_seq_length") and past_key_values.get_seq_length() == 0)
+        )
+        if is_prefill:
             model_inputs["pixel_values"] = pixel_values
 
         return model_inputs