[FEATURE]: Support keyless custom OpenAI-compatible providers

Feature hasn't been suggested before.\n\n- [x] I have verified this feature I'm about to request hasn't been suggested before.\n\n### Describe the enhancement you want to request\n\nAllow OpenCode’s native LLM runtime to use custom @ai-sdk/openai-compatible providers without requiring an API key.\n\nSelf-hosted inference services such as SGLang, vLLM, Ollama, and LM Studio are commonly configured as custom providers and may intentionally accept unauthenticated requests over a private network.\n\nThe native runtime currently rejects custom provider IDs and separately requires a non-empty API key. Together, these restrictions prevent valid self-hosted OpenAI-compatible endpoints from using the native runtime.\n\nDesired behavior:\n\n- Accept custom @ai-sdk/openai-compatible provider IDs in the native runtime.\n- Allow apiKey to be absent.\n- Send an authorization header only when a key is configured.\n- Preserve existing behavior for authenticated providers.\n\nRelated issues:\n\n- #38893 covers the native runtime’s hardcoded provider allowlist.\n- #39072 covers unconditional authorization requirements for another provider.\n\nThis request covers the combined custom-provider and optional-authentication behavior needed by self-hosted OpenAI-compatible endpoints.