LLM-Rubric: A Multidimensional, Calibrated Approach to Automated Evaluation of Natural Language Texts