Query generators for RAG query

[hardikjshah]

Sometimes when doing RAG, we need custom logic to convert the user messages to a query that can be sent to memory banks for retrieval purposes. This could be as simple as strong concatenation in some cases or using an LLM for query generation.

This PR introduces a notion of MemoryQueryGenerators and associated configs. Agent reads this config to understand how to generate the query before calling memory apis.

Tested with --

1. Start distro

   llama stack run local --name {distro_name} 

2. Run RAG script

   PYTHONPATH=. python examples/scripts/rag_with_memory_bank.py localhost 5000 --disable-safety 

[hardikjshah]

Done both.

TODOs:

• Rename to more standard name if possible
• LLM generator config should not take host / port