RAGtruth integration

[FrejaThoresen]

• Adds full RAGTruth processing pipeline, including new preprocessing and translation scripts for multilingual data preparation.
• Introduces an end-to-end ground-truth evaluation workflow with new modules and scripts for annotation, schema upgrades, hallucination dataset generation, LLM judging, and evaluation against human labels.
• Expands core package capabilities with new evaluation and logging components, plus updates to dataset generation, model generation, training, prompt utilities, and hallucination detection.
• Adds baseline and utility script improvements for generation, detection, and detector training workflows.
• Updates prompt templates across many languages and adds a new Albanian prompt file.
• Strengthens testing by adding ground-truth evaluation tests.
• Refreshes project docs and analysis artifacts, including new research/annotation docs and human evaluation outputs.
  Updates developer and runtime tooling (Dockerfile, docker-compose, pyproject, lockfile, makefile, pre-commit, mise, gitignore).