🐛 Fix(Gemini LLM): Support Gemini 0.2.x plugin on agent app

In `base_app_runner.py` and `easy_ui_based_generate_task_pipeline.py`,
modified to handle cases where `delta.message.content` is not only a string
but also a list containing objects like `TextPromptMessageContent`.
This allows supporting diverse LLM output formats.

The previous code assumed that LLM responses would always be a single string. However, multimodal models or models generating more complex responses might return content as a list of objects, such as `TextPromptMessageContent`. This change allows the application to correctly process and display these new forms of responses to the user. This enhances support for future LLM feature extensions.

fixes #20204

api/core/app/apps/base_app_runner.py

@@ -317,7 +317,14 @@ class AppRunner:
             else:
                 queue_manager.publish(QueueAgentMessageEvent(chunk=result), PublishFrom.APPLICATION_MANAGER)
 
+            if isinstance(result.delta.message.content, str):
                 text += result.delta.message.content
+            elif isinstance(result.delta.message.content, list):
+                for content in result.delta.message.content:
+                    if not isinstance(content, str):
+                        text += content.data
+                    else:
+                        text += str(content)  # failback to str
 
             if not model:
                 model = result.model

api/core/app/task_pipeline/easy_ui_based_generate_task_pipeline.py

@@ -48,6 +48,7 @@ from core.model_manager import ModelInstance
 from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk, LLMResultChunkDelta, LLMUsage
 from core.model_runtime.entities.message_entities import (
     AssistantPromptMessage,
+    TextPromptMessageContent,
 )
 from core.model_runtime.model_providers.__base.large_language_model import LargeLanguageModel
 from core.ops.entities.trace_entity import TraceTaskName
@@ -309,6 +310,23 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline):
                 delta_text = chunk.delta.message.content
                 if delta_text is None:
                     continue
+                if isinstance(chunk.delta.message.content, list):
+                    delta_text = ""
+                    for content in chunk.delta.message.content:
+                        logger.debug(
+                            "The content type %s in LLM chunk delta message content.: %r", type(content), content
+                        )
+                        if isinstance(content, TextPromptMessageContent):
+                            delta_text += content.data
+                        elif isinstance(content, str):
+                            delta_text += content  # failback to str
+                        else:
+                            logger.warning(
+                                "Unsupported content type %s in LLM chunk delta message content.: %r",
+                                type(content),
+                                content,
+                            )
+                            continue
 
                 if not self._task_state.llm_result.prompt_messages:
                     self._task_state.llm_result.prompt_messages = chunk.prompt_messages