Support CFORGE_DEFAULT_MODEL env var for configurable default model

[ermin-muratovic]

Summary

Allow users to configure the default LLM provider and model via environment variables.

Depends on

cognitive-forge #16 — pluggable LLM provider architecture must be implemented first.

Current Behavior

All commands default to OpenRouter + openai/gpt-4o hardcoded in source.

Desired Behavior

# Use Claude directly (no OpenRouter needed)
CFORGE_PROVIDER=claude
ANTHROPIC_API_KEY=sk-ant-...

# Use Ollama locally (zero cost, data stays local)
CFORGE_PROVIDER=ollama
CFORGE_DEFAULT_MODEL=llama3

# Use OpenRouter (current default, unchanged)
CFORGE_PROVIDER=openrouter
OPENROUTER_API_KEY=sk-or-...
CFORGE_DEFAULT_MODEL=openai/gpt-4o

Implementation

After cognitive-forge #16 lands:

• CFORGE_PROVIDER env var selects the provider (claude/ollama/openrouter/custom)
• CFORGE_DEFAULT_MODEL selects the model within that provider
• Provider factory in infrastructure layer resolves the correct LLMClient
• /model <name> in chat still overrides per-session
• .env.example updated with all provider options
• README documents provider configuration

Acceptance Criteria

• cognitive-forge Add /milestone command to cforge-dev chat #16 merged and available
• CFORGE_PROVIDER selects provider across all commands
• CFORGE_DEFAULT_MODEL sets model within provider
• Falls back to openrouter + gpt-4o if neither set (no breaking change)
• .env.example updated
• README documents all provider options