Add gemini

[wongjingping]

Changes

• Add gemini client. Some caveats:
  • Google's client has a weird thing where not only the system instructions are passed in separately during the GenerativeModel, but the latest message and the history (excluding the latest message) are passed separately. The history goes into start_chat while the latest message goes into send_message, which is really weird imho.
  • Unfortunately the seed parameter is not supported in the latest version: GenerationConfig Error: 'seed' Field Missing in Protocol Message but Present in GenerationConfig Class google-gemini/generative-ai-python#605, so we just made a short note.
• Add multi-llm querying in parallel. This is mostly for agents-style workloads.

Added tests for the gemini client and multi-llm querying. Have also added GEMINI_API_KEY into our organization secrets to facilitate CI testing.

[Muhammad18557]

Amazing, thank you and +111 for the tests like always:)

API signature seems to be largely similar to other LLMS but wondering why they do not support a seed parameter!