Add native Ollama provider support

[ianu82]

Summary

• add a native ollama LLM provider using Ollama's Python client and /api/chat
• add Ollama as a first-class setup/onboarding option with local model discovery
• wire Ollama through scratchpad, streaming reasoning progress, and provider-scoped verifier handling

What changed

• added anton/llm/ollama.py for native Ollama inference and model listing
• extracted shared OpenAI-style message/tool translation into anton/llm/openai_compat.py
• added shared LLM setup helpers in anton/llm/setup.py and reused them from CLI setup and /setup
• added ANTON_OLLAMA_BASE_URL settings support and scratchpad Ollama propagation
• kept Anthropic and OpenAI provider payload logic unchanged while widening the provider interface for Ollama-only request options

Verification

• pytest -q (411 passed)
• live native Ollama smoke checks against local models
  • qwen3:0.6b streamed reasoning progress through the new Ollama path
  • ministral-3:8b returned normal native Ollama responses
  • ChatSession.turn(...) completed successfully with Ollama configured as both planning and coding provider