Feature: support for collection-based windowed aggregations

[daniil-quix]

Add support for windowed aggregations that return collections (aka "aggregate()") rather than a single reduced value ("reduce()").
This type of API is helpful for aggregations requiring full datasets (e.g., medians), data buffering for ML models, etc.

Currently, aggregating collections is technically possible with reduce(), but it's not optimal:

• Every time a new record comes to the window, the whole window state needs to be deserialized, updated, and serialized back.
• Due to the above, more data will be sent to the changelog topic on every update, which may overflow the max message size in Kafka.

The API examples will be added later to this issue as research continues.