A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?