feat: Add OpenAI GPT adapter for multi-provider LLM support

[Copilot]

CAIMS was hardcoded to Anthropic Claude. This adds an OpenAI adapter and a provider factory so deployments can switch between providers via environment variable.

New files

• lib/adapters/openai.ts — OpenAIAdapter implementing the same LLMAdapter interface (chat() + judge()). Uses native fetch against the OpenAI chat completions API with exponential backoff retries. Supports OPENAI_BASE_URL for OpenAI-compatible endpoints.
• lib/adapters/index.ts — getAdapter() factory reads CAIMS_LLM_PROVIDER env var, returns the appropriate singleton. Defaults to anthropic.

Provider selection

# .env
CAIMS_LLM_PROVIDER=openai        # or anthropic (default)
OPENAI_API_KEY=sk-...
OPENAI_BASE_URL=https://api.openai.com/v1   # optional, for compatible APIs

The existing chat route already calls getAdapter() — no route changes needed.

Tests

• __tests__/openai.test.ts — 11 tests: response mapping, request shape, custom base URL, missing key, empty content, API errors
• __tests__/adapter-factory.test.ts — 8 tests: default provider, env var parsing (case/whitespace insensitive), explicit override, unknown provider error