Fix aggregator batching for nullable aggregates

[folmate]

Problem:
Aggregators such as min/max have a null initial value, resulting in changes not being propagated.
(The entry set of oldValues discards keys, where the value is null.)

Solution:
Initial values are wrapped in an Optional, thus null values will not be directly stored in oldValues.

[kris7t]

Thanks for the contribution!

To be honest, I wouldn't want to add one more layer of indirection (Optional<T>) to aggregators. In the common case, the aggregate result is never null, since aggregators act as outer joins. I updated to the code to clarify this with @NotNull annotations to hopefully make the API semantics more clear for users: ac4f28d

I'd be willing to add another, "nullable" aggregator that can entirely remove matches. This should ideally map to a different RETE node (with extra indirection if needed). However, this isn't a common use-case for Refinery, as entirely removing an aggregator match makes abstract interpreters quite hard to define.