Creativity or Brute Force? Using Brainteasers as a Window into the Problem-Solving Abilities of Large Language Models