Remove Legacy WatsonX Implementation

[manavgup]

Problem

The codebase currently has two WatsonX implementations:

1. Legacy: backend/vectordbs/utils/watsonx.py - Standalone utility functions
2. Modern: backend/rag_solution/generation/providers/watsonx.py - Factory pattern implementation

Analysis

The modern version (rag_solution/generation/providers/watsonx.py) is actively used in core services:

• PipelineService uses LLMProviderFactory.get_provider()
• QuestionService uses the provider factory pattern
• SystemInitializationService initializes WatsonX models

The legacy version is still imported in 14+ files but not used in core application logic.

Files Using Legacy Implementation

Vector Store Implementations (5 files):

• backend/vectordbs/weaviate_store.py
• backend/vectordbs/chroma_store.py
• backend/vectordbs/milvus_store.py
• backend/vectordbs/elasticsearch_store.py
• backend/vectordbs/pinecone_store.py

Data Processing (3 files):

• backend/rag_solution/data_ingestion/pdf_processor.py
• backend/rag_solution/data_ingestion/chunking.py
• backend/rag_solution/doc_utils.py

Evaluation & Analysis (3 files):

• backend/rag_solution/evaluation/evaluator.py
• backend/rag_solution/evaluation/llm_as_judge_evals.py
• backend/rag_solution/query_rewriting/query_rewriter.py

Tests & Documentation (3 files):

• backend/tests/unit/test_settings_dependency_injection.py
• backend/SETTINGS_MIGRATION_PLAN.md
• docs/fixes/TEST_ISOLATION.md

Solution

1. Update all imports from vectordbs.utils.watsonx to use LLMProviderFactory
2. Replace direct function calls with provider pattern:
   • get_embeddings() → provider.get_embeddings()
   • generate_text() → provider.generate_text()
3. Update vector stores to use provider factory for embeddings
4. Update data processing to use provider pattern for text generation
5. Remove legacy file backend/vectordbs/utils/watsonx.py

Benefits

• ✅ Consistent architecture - all LLM interactions use the same provider pattern
• ✅ Better testability - providers can be easily mocked via factory
• ✅ Database-driven configuration - no hardcoded settings dependencies
• ✅ Resource management - proper client lifecycle management
• ✅ Error handling - standardized LLM provider error handling

Priority

High - This cleanup will improve code consistency and maintainability.

Acceptance Criteria

• All 14 files updated to use modern provider pattern
• Legacy vectordbs/utils/watsonx.py file removed
• All tests pass
• No breaking changes to existing functionality
• Documentation updated to reflect new patterns