As long the invariants are not violated, the operation is technically valid. The insertion algorithm in CLRS splits on the way down, so it would split the root like you did.
However, another implementation might observe that the second child is empty and the first is full. That means the "rotation" can be done and the root node count is unaffected. The rotation involves pushing L down into the second child (prepending) and pulling up I up into L's previous place in the root. Now the first child has only two entries and you can insert into it.