Fix Ollama adapter to handle system messages properly

[Maximooch]

This PR fixes issue #4 by modifying the OllamaAdapter class to properly handle system messages.

Changes:

1. Updated the format_messages method in OllamaAdapter to convert system messages to user messages with a special prefix
2. Enhanced the process_response method to handle different response formats from Ollama
3. Updated the supports_system_messages method to return False to indicate native support is missing
4. Added documentation for Ollama configuration in the docs
5. Added tests to verify the adapter works correctly

These changes allow Penguin to work properly with Ollama models by ensuring system prompts and instructions are still passed to the model effectively.

[Maximooch]

Fixes #4

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
penguin	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Mar 31, 2025 6:21pm

[Maximooch]

I (OpenHands an AI Agent, not Maximooch a Human) have added additional fixes to this PR:

1. Fixed ModelConfig handling in core.py to properly handle both dictionary-like and object-like config.model
2. Made the Perplexity API key optional to avoid errors when running with Ollama
3. Updated the config.yml to use Ollama with the deepseek-r1:1.5b model

These changes allow Penguin to run properly with Ollama models, and the system messages are correctly converted to user messages with a special prefix as shown in the logs: "Converting system message to user message for provider: ollama".