Feat/deepseek adapter dropin 14224636701039833263 (#50)

* fix: resolve 422 error adding deepseek provider

- Updated `pretor/api/provider.py` to allow "deepseek" as a valid Literal in `ProviderRegister` Pydantic model.
- Validated tests to ensure the backend can correctly receive deepseek configurations.

Co-authored-by: zhaoxi826 <198742034+zhaoxi826@users.noreply.github.com>

* fix: complete deepseek provider registration wiring

- Updated `pretor/core/global_state_machine/provider_manager.py` to correctly map `"deepseek"` to `DeepseekProvider`.
- Updated `pretor/core/global_state_machine/model_provider/__init__.py` to export `DeepseekProvider`.
- Confirmed this fully resolves the Provider Manager failing to instantiate DeepSeek despite passing API validation.

Co-authored-by: zhaoxi826 <198742034+zhaoxi826@users.noreply.github.com>

* fix: support pydantic-ai decorator proxying on DeepSeekReasonerAgent

- Implemented `__getattr__` on `DeepSeekReasonerAgent` to safely proxy all unrecognized attributes (such as `@agent.system_prompt` and `@agent.tool`) directly to the underlying PydanticAI `Agent` object.
- Resolves the crash where `SupervisoryNode.create_agent()` threw an `AttributeError` when trying to decorate `system_prompt`.

Co-authored-by: zhaoxi826 <198742034+zhaoxi826@users.noreply.github.com>

* refactor: remove gemini provider from frontend and backend

- Removed `gemini` from `ProviderRegister` API validator.
- Removed `GeminiProvider` files, tests, and its mappings from `AgentFactory` and `ProviderManager`.
- Removed `gemini` from frontend TypeScript types and UI selection dropdown.

Co-authored-by: zhaoxi826 <198742034+zhaoxi826@users.noreply.github.com>

* fix: parse output with TypeAdapter to support Union types

- Refactored `_parse_output` in `DeepSeekReasonerAgent` to use Pydantic's `TypeAdapter`.
- Resolves a bug where Union types (like `Union[ForConsciousnessNode, ForUser]`) evaluated `hasattr(..., 'model_validate_json')` as `False`, causing the parser to fall back to `json.loads` and return a raw `dict`.
- The `SupervisoryNode` now correctly receives Pydantic objects instead of dictionaries, resolving the `未知响应类型: <class 'dict'>` crash.
- Cleaned up debug scripts to adhere to repository standards.

Co-authored-by: zhaoxi826 <198742034+zhaoxi826@users.noreply.github.com>

---------

Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Co-authored-by: zhaoxi826 <198742034+zhaoxi826@users.noreply.github.com>

pretor/adapter/model_adapter/deepseek_reasoner.py

@@ -7,6 +7,17 @@ from pydantic_ai.run import AgentRunResult
 
 T = TypeVar('T', bound=BaseModel)
 
+class AgentRunResultProxy:
+    def __init__(self, original, parsed):
+        self._original = original
+        self._parsed = parsed
+    def __getattr__(self, name):
+        if name == 'data':
+            return self._parsed
+        if name == 'output':
+            return self._parsed
+        return getattr(self._original, name)
+
 class DeepSeekReasonerAgent(Generic[T]):
     """
     专为 DeepSeek-V4/R1 设计的适配器。
@@ -32,12 +43,9 @@ class DeepSeekReasonerAgent(Generic[T]):
         format_instruction = ""
         if self.has_custom_output:
             try:
-                if hasattr(self.output_schema, 'model_json_schema'):
-                    schema_str = json.dumps(self.output_schema.model_json_schema(), ensure_ascii=False)
-                else:
-                    # Don't inject <class 'dict'> into prompt
-                    schema_str = str(getattr(self.output_schema, '__name__', str(self.output_schema)))
-
+                from pydantic import TypeAdapter
+                schema_dict = TypeAdapter(self.output_schema).json_schema()
+                schema_str = json.dumps(schema_dict, ensure_ascii=False)
                 format_instruction = (
                     f"\n\nCRITICAL: 你必须输出且只能输出一段纯 JSON 格式的数据，"
                     f"并包裹在 ```json 和 ``` 之间。格式必须符合以下 JSON Schema 结构（或对应数据类型）：\n"
@@ -95,10 +103,9 @@ class DeepSeekReasonerAgent(Generic[T]):
             raise ValueError("未找到有效的 JSON 块。请将结果包装在 ```json 中。")
 
         try:
-            if hasattr(self.output_schema, 'model_validate_json'):
-                return self.output_schema.model_validate_json(json_str)
-            else:
-                return json.loads(json_str)
+            from pydantic import TypeAdapter
+            adapter = TypeAdapter(self.output_schema)
+            return adapter.validate_json(json_str)
         except ValidationError as e:
             raise ValueError(f"返回的 JSON 无法匹配所需结构：{e}")
         except json.JSONDecodeError as e:
@@ -128,17 +135,6 @@ class DeepSeekReasonerAgent(Generic[T]):
                 parsed_data = self._parse_output(raw_text)
 
                 # Proxy the result to inject the parsed data seamlessly
-                class AgentRunResultProxy:
-                    def __init__(self, original, parsed):
-                        self._original = original
-                        self._parsed = parsed
-                    def __getattr__(self, name):
-                        if name == 'data':
-                            return self._parsed
-                        if name == 'output':
-                            return self._parsed
-                        return getattr(self._original, name)
-
                 return AgentRunResultProxy(result, parsed_data)
 
             except ValueError as e: