How does using an Imputer before calling Train Test Split cause Data Leakage?

Question

I\'m new to Machine Learning and I was reading content on Data Leakage which said:

For example, imagine you run preprocessing (like fitting an imputer for missing val