Quote:
Why there is a rule in first normal form that all underlying domain contain atomic value?
Because the goal of 1NF is that the same as the goal of all normalisation: to prevent the possibility of redundant data that can become out of sync (because if you store a customer address in two places, when they move premises it's very easy to update just one of them for example) and prepare for future scalability.
While it's easy to store a list of telephone numbers (say) in a comma separated list as a single column, it's also very easy to duplicate those numbers, because the field is just text and contains multiple values. 1NF requires each column to be atomic: it can hold one
and only one value, including the elimination of what are called "repeating groups" such as that list of telephone numbers. By storing each telephone number in its own atomic column (and / or row) and relating that back to the customer that it applies to, you end you with multiple references to the same row, which when updated automatically updates all the referencing data.