Build a module that suggests alt text for images in educational content. Since we can't run image recognition locally, focus on context-based suggestions:
- Parse surrounding text (caption, paragraph, figure element) to infer image purpose
- Detect common educational image patterns from filenames and paths (diagram, chart, graph, equation, photo)
- Generate descriptive alt text templates: "[Type] showing [inferred subject].[context from surrounding text]"
- Flag images that likely need human review (complex diagrams, photos)
- Handle decorative images (suggest empty alt="")
Input: HTML string. Output: Array<{ imgSrc: string, currentAlt: string, suggestedAlt: string, confidence: number, needsReview: boolean }>
Write tests with realistic educational HTML samples.