fix: make onboarding not loop when error occurs in ingestion, fix double tasks for OpenRAG docs

[lucaseduoli]

This pull request introduces significant improvements to the onboarding rollback process and document ingestion logic, with a focus on enhanced cleanup, configurability, and consistency in connector types. The main changes include allowing partial rollback of onboarding (embedding-only), ensuring knowledge filters are cleaned up during rollback, purging tasks more thoroughly, and standardizing connector types for document ingestion tasks. Additionally, the ingestion pipeline now consistently returns task IDs and supports reusing existing tasks.

Onboarding Rollback Improvements:

• Added support for partial onboarding rollback (embedding-only) by introducing an embedding_only parameter in both the frontend (RollbackParams) and backend (RollbackBody), allowing users to roll back only the embedding step without resetting all onboarding progress. [1] [2] [3]
• Enhanced backend rollback logic to clean up knowledge filters created during onboarding and to fully purge tasks from memory for both authenticated and anonymous users, improving system hygiene and preventing stale data. [1] [2]

Frontend Integration:

• Updated all calls to the rollback mutation in the onboarding UI to pass the embedding_only flag, ensuring the correct rollback mode is triggered based on user context. [1] [2]

Document Ingestion Pipeline Enhancements:

• Refactored ingestion functions to consistently return task IDs, enabling better tracking and chaining of ingestion steps. Added support for passing existing_task_id through the pipeline to reuse tasks when appropriate. [1] [2] [3] [4] [5] [6] [7] [8]
• Standardized the connector_type used for document ingestion tasks to "openrag_docs" instead of "system_default" or "local", improving traceability and consistency across different ingestion methods. [1] [2] [3] [4] [5]

Other Improvements:

• Minor UI logic fixes and code formatting for onboarding steps and error handling.
• Ensured that error toasts for failed tasks are only shown if onboarding is not active, preventing redundant or confusing notifications.

These changes collectively make the onboarding and document ingestion flows more robust, maintainable, and user-friendly.

Closes #1243

[mpawlow]

@lucaseduoli

Code Review 1

• ✅ LGTM / Approved
• Approved based on code review for duplicate PR already merged into main branch
  • #1244