feat: Multi-provider LLM support (Ollama, OpenAI, Claude, Gemini)

[euvhmac]

What

Adds support for multiple LLM providers beyond Ollama — users can now run METATRON with OpenAI (GPT), Anthropic (Claude), Google (Gemini), or keep using Ollama locally. Provider switching is available both at startup and mid-session via a new Settings menu.

Why

Ollama is great for local/offline use, but not everyone has the hardware to run large models. Cloud providers give users access to stronger models with zero setup. This also opens the door for comparing analysis quality across different LLMs.

How

New files

File	Purpose
config.py	Centralized configuration loaded from .env via python-dotenv. Defines active provider/model, API keys, DB credentials, and provider catalog.
providers.py	Provider abstraction layer. BaseProvider ABC with send(messages) and ping() methods. Concrete implementations: OllamaProvider, AnthropicProvider, OpenAIProvider, GoogleProvider. Factory function get_provider().
.env.example	Template with all configuration variables and safe defaults.

Modified files

File	Change
llm.py	ask_ollama() → ask_llm() using the provider abstraction. summarize_tool_output() updated to use active provider. Preserves all upstream features (evidence gating, _clean(), summarizer, parser).
metatron.py	New Settings menu (view providers, switch active, test connection). Provider selection prompt before each scan. Banner shows active provider/model dynamically. Cross-platform terminal clear.
db.py	Uses config.py for DB credentials instead of hardcoded values.
export.py	Uses config.py for DB credentials instead of hardcoded values.
requirements.txt	Added python-dotenv, anthropic, openai, google-genai.
.gitignore	Added .env to prevent accidental API key leaks.

Architecture

User → metatron.py → llm.py → providers.py → [Ollama | OpenAI | Claude | Gemini]
↑
config.py (.env)

All providers implement the same BaseProvider interface, so llm.py is completely provider-agnostic. The system prompt, evidence gating rules, tool pipeline, and report generation work identically regardless of which provider is active.

Backward Compatibility

• Ollama remains the default — zero config change needed for existing users
• No database schema changes
• No changes to the tool pipeline or scan logic
• .env file is optional; defaults match current behavior
• Cloud SDKs are optional runtime dependencies (only imported when the provider is selected)

Setup (for cloud providers)

cp .env.example .env
# Edit .env and add your API key(s):
# ANTHROPIC_API_KEY=sk-ant-...
# OPENAI_API_KEY=sk-...
# GOOGLE_API_KEY=AIza...