The list of ID's passed into the chunker must be an instance of the struct BatchProcessingIDList. In general, it's possible to construct a modeled entity operation that returns an instance of this struct or populate an instance of this struct using the results of one or more entity operations in code.
However, if necessary, complex business logic could be constructed to populate this list with an appropriate set of IDs. But, in this instance, it's important to consider that this processing will all take place in the chunkers thread of execution and it may be more efficient to forgo some of the optimization at this point and allow the streams to filter out some instances where no work is required as this effort will be parallelized across all the streams.
While the infrastructure assumes that the IDs passed in this struct are single keys, it is possible to use composite keys if necessary, see Using Composite Keys to Identify Processing Units.