Improve AI tagging accuracy and training tools

Problem
AI tagging accuracy cannot be fine-tuned — admins have no way to provide negative reinforcement, flag good examples, or audit tag predictions, and the AI mispredicts on signatures and forwarded content.

Request
Add training and feedback tools for AI tagging: negative/positive reinforcement, sample auditing, training on more than 10 tags, and ignoring signatures and nested forwarded content.