Why Tfidf vector magnitude should be 1?

Question

I will be happy to understand why Tfidf vector magnitude should be 1. I know that instead of cosine_similarity one can use linear_kernel to reduce calc