The SIFo Benchmark: Investigating the Sequential Instruction Following Ability of Large Language Models