GLUECoS: An Evaluation Benchmark for Code-Switched NLP