Fix server for openai package

[awni]

1. Launch server:

mlx_lm.server

2. Make request:

import openai

openai_client = openai.OpenAI(api_key = "placeholder-api", base_url="http://localhost:8080")

response = openai_client.chat.completions.create(
    model='mlx-community/Meta-Llama-3-8B-Instruct-4bit',
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": f"Say this is a test!"},
    ],
)

Closes #875

Also added a couple fixes for RecurrentGemma 9B

[angeloskath]

🚀