fix: DPO tool role KeyError (#3217), dataset hash output_dir (#3303), config validators (#3538) [skip ci]

* fix: DPO tool role KeyError, dataset hash output_dir, config validators [skip-e2e]

- Add 'tool' to default role_map_inv in dpo/chat_template.py default() and
  argilla_chat() so datasets with tool-call messages no longer raise
  KeyError: 'tool' (closes #3217)

- Fix generate_dataset_hash_from_config to use canonical tokenizer config +
  overrides content instead of tokenizer.name_or_path when added_tokens_overrides
  is set, preventing cache busting when only output_dir changes (closes #3303)

- Add three Pydantic config validators to AxolotlConfigWCapabilities:
  * save_strategy: 'best' requires metric_for_best_model
  * streaming=True is incompatible with val_set_size > 0
  * lora_target_modules list entries must be valid Python regex patterns

- Tests for all three changes

* review: condense comment in shared.py, swap Mistral model for SmolLM2-135M in test_hash

* chore: lint

* move the validators out of the w/ capabilities schema

---------

Co-authored-by: Wing Lian <wing@axolotl.ai>

tests/prompt_strategies/test_dpo_chat_templates.py

@@ -294,5 +294,88 @@ class TestArgillaChatDPOChatTemplate:
         assert result["rejected"] == "party on<|end|>"
 
 
+class TestDPOChatTemplateToolRole:
+    """
+    Test that DPO chat template strategy handles tool role messages without KeyError.
+    Regression test for https://github.com/axolotl-ai-cloud/axolotl/issues/3217
+    """
+
+    def test_tool_role_default_no_key_error(self, llama3_tokenizer):
+        """Messages list with a 'tool' role should not raise KeyError."""
+        dataset = Dataset.from_list(
+            [
+                {
+                    "messages": [
+                        {"role": "user", "content": "What is the weather?"},
+                        {
+                            "role": "assistant",
+                            "content": "Let me check.",
+                        },
+                        {
+                            "role": "tool",
+                            "content": "22°C, sunny.",
+                        },
+                    ],
+                    "chosen": {
+                        "role": "assistant",
+                        "content": "It is 22°C and sunny.",
+                    },
+                    "rejected": {
+                        "role": "assistant",
+                        "content": "I don't know.",
+                    },
+                }
+            ]
+        )
+        transform_fn, _ = default(
+            DictDefault(
+                {
+                    "chat_template": "llama3",
+                    "datasets": [{"type": "chat_template"}],
+                }
+            )
+        )
+        # Should not raise KeyError: 'tool'
+        result = transform_fn(dataset[0], tokenizer=llama3_tokenizer)
+        assert "prompt" in result
+        assert "chosen" in result
+        assert "rejected" in result
+
+    def test_tool_role_custom_mapping_preserved(self, llama3_tokenizer):
+        """A user-supplied roles mapping that overrides 'tool' is still respected."""
+        dataset = Dataset.from_list(
+            [
+                {
+                    "messages": [
+                        {"role": "user", "content": "hello"},
+                        {"role": "tool_result", "content": "42"},
+                    ],
+                    "chosen": {"role": "assistant", "content": "The answer is 42."},
+                    "rejected": {"role": "assistant", "content": "Unknown."},
+                }
+            ]
+        )
+        transform_fn, _ = default(
+            DictDefault(
+                {
+                    "chat_template": "llama3",
+                    "datasets": [
+                        {
+                            "type": "chat_template",
+                            "roles": {
+                                "user": ["user"],
+                                "assistant": ["assistant"],
+                                "system": ["system"],
+                                "tool": ["tool_result"],
+                            },
+                        }
+                    ],
+                }
+            )
+        )
+        result = transform_fn(dataset[0], tokenizer=llama3_tokenizer)
+        assert "prompt" in result
+
+
 if __name__ == "__main__":
     unittest.main()