[Enhancement]: Add robust error handling and retry logic to CouchbaseVectorStore search operations

[Igosain08]

Currently, CouchbaseVectorStore makes direct calls to the Couchbase Search service without internal exception handling. In production environments, transient issues (e.g., AmbiguousTimeoutException) or state issues (e.g., IndexNotFoundException during a rebalance) cause the entire RAG chain to fail.

Proposed Solution

I suggest wrapping the internal _search calls within similarity_search_with_score_by_vector with a custom retry/exception handler.

Logic Flow:

1. Try the search operation.
2. Except AmbiguousTimeoutException, ServiceUnavailableException: Implement a simple exponential backoff retry (e.g., 3 attempts).
3. Except IndexNotFoundException:
   • Log a logger.warning stating the index is not yet ready.
   • Return an empty list [] to allow the RAG chain to continue without crashing.
4. Except CouchbaseException: Log the specific internal error and re-raise as a standard RuntimeError (or a LangChain-native exception). This ensures that Couchbase-specific SDK details do not leak into the application layer for developers using the general VectorStore interface.