Refactor: Improve audio tagging and analysis

Commit notes

This commit enhances audio analysis by integrating AI-powered tagging. It refactors the `categorizeByRules` function to focus solely on category detection and introduces a new `ai-tagging` service. This service leverages LLMs for richer semantic tagging, mood, texture, and character descriptions. A fallback rule-based tagging mechanism is also included for scenarios where AI analysis is unavailable. The worker process is updated to utilize these new services, combining AI-generated tags with existing metadata for comprehensive sample analysis.

Co-authored-by: armin.naimi <[email redacted]>