Revisiting Metric Reliability for Fine-grained Evaluation of Machine Translation and Summarization in Indian Languages