Can we process queries offline? I think we can make Mo's algorithm work here.

Provide the problem link. I will post my solution.

TL;DR There exist variation of Mo's algorithm with only adds and rollbacks.

I assume that you understand solution with Mo and DSU ideas of how to use Mo and how to use DSU to solve one query in $$$O(N \alpha(N))$$$.

This is not Mo, but similar: Queries with length less then $$$\sqrt{N}$$$ we can answer in $$$O(len)$$$ time similar to what I will do next, let's leave it as an exercise. All longer queries we will group by the bucket (of size $$$\sqrt{N}$$$) in which its left border is. For one bucket we will sort all queries by right border. Let's say that borders of bucket are $$$L$$$ and $$$R$$$ (which means left borders of all queries are in $$$[L, R]$$$). We will start with segment $$$[R, R]$$$ and then will move right border to the right (adding the right element). When we encounter right border for some query, we will move left border to the left so that it coincides with left border of query. Then we will answer the query and rollback all the left border movements. The complexity is like in Mo's algorithm, the difference is that we do only adds and rollbacks.

Underlying DS is not DSU, but just remembering left border of segment for right border and vice versa. All the changes are in strict $$$O(1)$$$ time, so total complexity is $$$O(Q\sqrt{N})$$$.