IMO, it makes way more sense to decide the aggregations you want ahead of time (e.g. "SELECT customer, sum(value) FROM purchases GROUP BY customer"). That way, you deal with substantially less data and everything becomes a whole lot simpler.
What some would do is record in blocks where every point after the earliest is stored as a delta. Then each block is more compressible as it contains a lot of 0s.
Some tasks actually require absolute granularity, up to 6 decimal places of precision and thereafter reliance on atomic order of arrival, for deterministic results on data from high frequency trading.
Without absolute knowledge of the order or if there's aggregation the best you can do is approximate, which often is considered suboptimal when the real solution is available.
What I am doing with one program could almost be called a data lake because it is just a bunch of JSONL files that have really varied data in them. But it's organized by date and hour per day as well as predefined keys, since I know I will need to query it that way.