Refactor pipeline mode handling with strategy pattern and harden retrieval imports

[cursor]

Summary

• Centralized runtime mode validation in src/config.py using a single _read_mode() path and explicit allowed mode sets.
• Added a pipeline Strategy + Factory in src/core/pipeline/mode_strategy.py and reused it from both ingestion entry points.
• Replaced duplicated pipeline_mode branching in src/workers/tasks/ingestion.py and src/core/saving/auto_kg.py with strategy dispatch.
• Removed duplicated model mode constants by reusing MODEL_MODES from config in src/core/instances.py.
• Hardened retrieve controller import behavior by deferring heavy imports (kg_agent, NER extractor) to function scope to reduce import-time failure points.
• Stabilized architecture test stubs to match the retrieve controller surface.

Design Patterns Applied

• Strategy: PipelineModeStrategy implementations for lightweight and accurate behavior.
• Factory Method: PipelineModeStrategyFactory.create() selects strategy by mode.

Tests

• Added tests/test_config_pipeline_mode_refactors.py covering:
  • mode validation and defaults (PIPELINE_MODE, OCR_MODE, AGENTIC_ARCHITECTURE)
  • strategy factory behavior and invalid mode rejection
  • integration assertions that ingestion and auto-kg use strategy dispatch
• Full suite run:
  • python3 -m pytest -q
  • Result: 52 passed, 1 warning

Commits

• 5dc8549 Refactor pipeline mode handling with strategy factory
• 4db81d7 Defer heavy retrieve imports and stabilize architecture stubs