feat: add OpenAI annotator support

[joryirving]

What

Add OpenAI annotator support to the Annotate step, matching the existing OpenAI filter support.

Closes #41

Why

Users running local inference servers (llama.cpp, text-generation-webui, etc.) can now use both filtering and annotation without needing a separate Ollama instance. This is especially useful for CPU-only deployments where Ollama's overhead is undesirable.

Changes

• New: annotator_openai.go — OpenAI annotator implementation (166 lines)
  • Uses openai-go client with configurable base URL
  • Regex-based JSON extraction from chat responses (compatible with any OpenAI-compatible server)
  • Same response format as Ollama annotator: model_feedback, processed_number, processed_text, question_answer
  • Supports API key, model selection, timeout, system prompt override
  • Default timeout 60s for slower CPU models
• Modified: config.go — added OpenAI OpenAIAnnotator field to AnnotateStep struct
• Modified: annotators.go — registered OpenAI annotator in the annotators slice

Example config

Steps:
- Filter:
    OpenAI:
      Model: qwen3.5-9b
      URL: http://llama.llm:8080/v1
      UserPrompt: >
        Should this message be forwarded? Look for human-written content...
- Annotate:
    OpenAI:
      Model: qwen3.5-9b
      URL: http://llama.llm:8080/v1
      UserPrompt: >
        Rate the emotional content of this message from pilots and ATC
        between 1-100 for any emotions such as anger, frustration, etc.

[tyzbit]

One minor question. Looks great to me though, thank you!