At Datadog we see over 30% of the time spent in `DisjointBuilder.add` is spent accessing the start and end members, and this is because `TimestampUnit.compareTo` calls `ScaleFactor.getMultiplier` which is generally a relatively expensive function (e.g. `Math.pow`). Restructuring the `IScalarAffineTransform` hierarchy so offsets, multipliers, and flags are stored as data will improve parsing efficiency.