Unmasking the Limits of Large Language Models: A Systematic Evaluation of Masked Text Processing Ability through MskQA and MskCal