Port POSet to Java

[Scott-Guest]

Part of #4030.

Port the POSet class from Scala to Java.

This is mostly a straightforward conversion from Scala functional idioms (map, filter, reduce, etc.) to the corresponding Java Stream methods. To port lazy vals, we also implement a Lazy<T> wrapper which caches the result of a Supplier<T>.

The only remaining reference to Scala here is a single constructor which takes a scala.collection.Set<Tuple2<T, T>> and internally converts it to a java.util.Set<Pair<T, T>>. This can be removed once we more pervasively switch away from Scala collection types everywhere in the codebase.

[Scott-Guest]

The output is unstable because it's based on the iteration order of relations.entrySet(), but I don't see a clean way to address this.

[Baltoli]

We can revisit this in the future if it becomes a worse problem; this code is pretty stable and so long as we aren't doing further big changes here we should be fine!

[gtrepta]

If we care more that there's a circularity, and less what the circularity is, we could sed out the example given in the error message.

[Baltoli]

I'm pleasantly surprised by how nicely the Java code comes out to be here!

[Baltoli]

Where does the need for Serializable come in? I presume you've thought this all through, just curious what the technical reason is!

[Scott-Guest]

The root of the issue is that lambdas and method references are not serializable by default. Instead, they are only serializable when constructed in a context whose target type is Serializable (https://stackoverflow.com/questions/25391656/serialization-of-a-lambda-after-its-creation).

In particular, this means that the Supplier<T> in each Lazy<T> in the POSet is not guaranteed to be serializable, giving errors when trying to serialize the overall POSet. We fix this by instead requiring SerializableSupplier<T>.

As for why the overall POSet needs to be serializable in the first place, this was admittedly error driven development - the original POSet class extends Serializable, and removing it causes tests to fail!

[Baltoli]

Looks good; I'm impressed by how readable the Java translation is and how minimal the API surface change is. Nice work!