fix: refactor Ollama model fetching to use async and filter capabilities (#10550)

* Refactor Ollama model fetching to use async and filter capabilities

Replaces synchronous requests for Ollama model fetching with asynchronous httpx calls and adds filtering to only include models with 'completion' capability. Updates the LanguageModelComponent to support async validation and fetching of Ollama models, improving reliability and accuracy of available model options.

* Update language_model.py

* Add Notion integration components

Introduces several Notion-related components for Langflow, including AddContentToPage, NotionDatabaseProperties, NotionListPages, NotionPageContent, NotionPageCreator, NotionPageUpdate, and NotionSearch. Updates the component index to register these new tools, enabling Notion API interactions such as page creation, content retrieval, database property listing, and more.

* [autofix.ci] apply automated fixes

* [autofix.ci] apply automated fixes (attempt 2/3)

* [autofix.ci] apply automated fixes (attempt 3/3)

* Update LanguageModelComponent code in starter projects

Refactored the LanguageModelComponent code in Basic Prompt Chaining.json to improve formatting and readability. No functional changes were made; only code style and structure were updated.

* Refactor Ollama model fetching logic

Moves Ollama model fetching and URL validation logic from the LanguageModelComponent class to shared utility functions in model_utils. Updates references in starter project JSONs to use the new utility functions for improved code reuse and maintainability.

* Update component_index.json

* Update Nvidia Remix.json

* Handle message state for streaming responses

Adds logic to skip processing text content when message state is 'complete' in OpenAI response streaming, ensuring only content_blocks are processed for tool calls. Updates LCModelComponent to set state to 'partial' during streaming and only update the database with 'complete' state, without sending a new message event, as the frontend already has all streamed content.

* fixed test

* Changed embedding model component

* removed default values

* fixed ruff

* fixed starter projects

---------

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: Lucas Oliveira <lucas.edu.oli@hotmail.com>

Author: 3 people

src/backend/base/langflow/api/v1/openai_responses.py

@@ -171,13 +171,30 @@ async def openai_stream_generator() -> AsyncGenerator[str, None]:
                                     sender = data.get("sender", "")
                                     content_blocks = data.get("content_blocks", [])
 
+                                    # Get message state from properties
+                                    properties = data.get("properties", {})
+                                    message_state = properties.get("state") if isinstance(properties, dict) else None
+
                                     await logger.adebug(
-                                        "[OpenAIResponses][stream] add_message: sender=%s sender_name=%s text_len=%d",
+                                        (
+                                            "[OpenAIResponses][stream] add_message: "
+                                            "sender=%s sender_name=%s text_len=%d state=%s"
+                                        ),
                                         sender,
                                         sender_name,
                                         len(text) if isinstance(text, str) else -1,
+                                        message_state,
                                     )
 
+                                    # Skip processing text content if state is "complete"
+                                    # All content has already been streamed via token events
+                                    if message_state == "complete":
+                                        await logger.adebug(
+                                            "[OpenAIResponses][stream] skipping add_message with state=complete"
+                                        )
+                                        # Still process content_blocks for tool calls, but skip text content
+                                        text = ""
+
                                     # Look for Agent Steps in content_blocks
                                     for block in content_blocks:
                                         if block.get("title") == "Agent Steps":