Hmmm, you can try understanding a node in the segment tree contains the information of all the numbers in the segment. When we want to combine two nodes, lets call them L and R, inversion of current node = inversion of L + inversion of R + inversion of array of (L, R). In order to do that, we can maintain a frequency table in each of the node (size 40). Also notice that it is better that we count that with a prefix sum in 40 operations, instead of doing a 40 * 40 operation (looping through i, j).

After merging the nodes, we still notice that our answer can't just simply sum all together, because the segment tree considers separated segments. But we can observe that the number of segments is around log N, so we can simply brute force them together to calculate the answer.

Code link here

the time complexity for my solution is $$$O(40 * N log N)$$$ which is passable.