Feat filter vectorstore

[gs-costa]

Add Filtering Capabilities to Vector Store Components

Overview

This PR adds filtering capabilities to all vector store components in Langflow, allowing users to narrow down search results based on specific field values. This enhancement improves search precision by enabling metadata-based filtering across all supported vector store backends.

Key Changes

• Added search_filter parameter to the base LCVectorStoreComponent class
• Implemented filter support in vector store components below:
  • Pinecone
  • Qdrant
  • Supabase
  • Milvus
  • FAISS
  • Elasticsearch
  • MongoDB (pre_filter)
  • Weaviate (where_filter)
  • Vectara
  • Upstash
• Updated the VectorStoreRAG template to support filtering
• Fixed filter handling in LocalDB and Elasticsearch
• filter support not implemented due to issues:
  • Clickhouse
  • OpenSearch
  • PGVector
  • Weaviate (need client connection fix)
  • Couchbase (do not allow filter)
  • Redis

Implementation Details

• Added NestedDictInput for search_filter in the base component
• Updated search_with_vector_store method to pass filter parameters to vector store search methods
• Modified each vector store component to properly handle the filter parameter
• Ensured consistent filter parameter naming across different backends

Usage Example

# Create a filter for a specific file path
search_filter = {
    "file_path": "/path/to/document.pdf"
}

# Use in vector store component
results = redis_vector_store.search_documents(
    search_query="your query",
    search_filter=search_filter
)

Testing

• Added tests for filter functionality across vector store components
• Verified filter behavior with different data types and filter expressions

Impact

Enables more precise search functionality by filtering results based on metadata, improving search relevance and performance across all supported vector store backends.