clue prompt templates

[yongzx]

Add prompt templates to the CLUE benchmark tasks.

Currently:

• 3 original task prompt templates
• 1 non-original task prompt template

WIP:

• NLI (ocnli, cmnli)
• IFLYTEK: it has 119 label names, and I haven’t populated the answer-choices yet.
• ChID: weird dataset structure where each example contains 4 subexamples. It needs the model to find 4 correct idioms for those 4 examples. The metrics are accuracy, and I am thinking of linearizing the dataset.