Auto-improve content extraction with OpenAI evaluation

[paskal]

Summary

• Adds automatic extraction quality evaluation using OpenAI during content extraction
• When OpenAI is configured (--openai-api-key) and no existing rule for the domain, GPT evaluates the extraction result and suggests CSS selectors if the result is poor
• Iterates up to 3 times (configurable via --openai-max-iter), saves the best selector as a rule for future use
• ExtractAndImprove() force mode ignores existing rules — for when a user reports bad extraction
• Protected POST /api/content-parsed-wrong?url=... endpoint for force mode
• Fail-open: GPT errors never break extraction — original result returned unchanged

Depends on #73 (modularise-retrieval).

Key design decisions

• Inline evaluation: runs during Extract(), not async. First request to a new domain may take longer (GPT calls), but always returns the best result. Subsequent requests use the saved rule
• GPT sees URL + extracted text + truncated HTML: enough context for accurate selector suggestions without excessive token cost
• Force mode: passes nil rule to general parser (ignores stored rules), then re-evaluates from scratch
• Image extraction deferred: extractPics (which downloads images via HTTP) runs once on the final result, not on every evaluation iteration

New configuration

Flag	Env	Default	Description
--openai-api-key	OPENAI_API_KEY	none	Enables auto-evaluation when set
--openai-model	OPENAI_MODEL	gpt-5.4-mini	Model for evaluation
--openai-max-iter	OPENAI_MAX_ITER	3	Max evaluation iterations

New interface

type AIEvaluator interface {
    Evaluate(ctx, url, extractedText, htmlBody, prevSelector string) (*EvalResult, error)
}