feat: support multiple LLM providers (OpenAI, Anthropic, Ollama)

[Aditya8840]

Summary

Closes #1

• Introduces a LLMProvider abstraction that decouples the agent loop from any specific LLM SDK
• Implements three providers: OpenAI, Anthropic (Claude), and Ollama (local models)
• Each provider manages its own message format, tool schema conversion, and API interaction behind a unified interface
• Adds --provider CLI flag and DROIDPILOT_PROVIDER env var for provider selection
• Anthropic SDK is an optional dependency: pip install droidpilot[anthropic]

Usage examples

# OpenAI (default, backward compatible)
droidpilot "Open Settings"

# Anthropic Claude
droidpilot "Open Settings" --provider anthropic --model claude-sonnet-4-20250514

# Local Ollama
droidpilot "Open Settings" --provider ollama --model llama3

Architecture

Component	Purpose
providers/base.py	LLMProvider ABC defining the interface
providers/openai_provider.py	OpenAI implementation
providers/anthropic_provider.py	Anthropic implementation with tool format conversion
providers/ollama_provider.py	Ollama via OpenAI-compatible API (inherits OpenAI provider)
providers/__init__.py	Registry, default models, create_provider() factory

Test plan

• 20 unit tests covering provider registry, tool format conversion, and all three provider implementations
• black . formatting passes
• mypy . passes (only pre-existing ui_tree.py error remains)
• Manual test with OpenAI provider
• Manual test with Anthropic provider
• Manual test with Ollama provider

🤖 Generated with Claude Code