fix: preserve metadata in RetrieverTool when using Custom Retriever

[limbergarcia]

🐛 Bug Description

When using Custom Retriever with Retriever Tool, metadata from vector store (e.g. Qdrant) is lost and only pageContent is preserved.

Root Cause

CustomRetriever overrides getRelevantDocuments() and returns formatted documents, but RetrieverTool internally calls retriever.invoke(), which uses LangChain's _getRelevantDocuments() instead.

This causes:

• Custom formatting to be ignored
• Metadata handling inconsistency

Fix

Updated CustomRetriever to ensure compatibility with LangChain's invoke() flow and preserve metadata correctly.

Steps to Reproduce

1. Use Qdrant vector store with metadata
2. Connect:
   Qdrant → CustomRetriever → RetrieverTool → Agent
3. Query documents
4. Metadata is missing in response

Expected Behavior

Metadata fields like:

• filename
• url
• blobType

should be preserved and accessible in the final output.

🔗 Related Issue

Closes #6242

[gemini-code-assist]

Code Review

This pull request refactors the CustomRetriever class by delegating document retrieval to a new _getRelevantDocuments method and ensuring resultFormat defaults to a standard value. A review comment correctly identifies a potential issue where special replacement patterns in pageContent could cause unexpected behavior during string replacement, suggesting a functional approach to ensure literal interpretation.