Scan/Reduce with Seed Factory

[benjchristensen]

Adds overload with seed factory as per #1831

[benjchristensen]

cc @headinthebox for review

[benjchristensen]

Merging for another release candidate to get actual testing during this next week.

[headinthebox]

For the mutable case, I think it is better to pass a special object such that we can hide the state, and all methods can access it without passing it as a parameter. Think something along the lines of http://msdn.microsoft.com/en-us/library/ms131051.aspx

[benjchristensen]

A special object means we can't use lambdas. The point of scan/reduce for most use cases I see is for accumulating state in a mutable object, so not supporting use of lambdas for the mutable case effectively says scan would require using an anonymous inner class. That seems unnecessary when I can write code like this:

        Observable.range(0, 1000).reduce(() -> new ArrayList<Integer>(), (list, i) -> {
            list.add(i);
            return list;
        }).forEach(System.out::println);

[headinthebox]

Yup, it would require an anonymous inner class, but it is also what Java 8 does https://docs.oracle.com/javase/8/docs/api/java/util/stream/Collector.html, https://docs.oracle.com/javase/8/docs/api/java/util/stream/Collectors.html.

[benjchristensen]

I'm okay if that's an extra overload, but I wouldn't want it to be the only way. It's very awkward using an anonymous inner class when nothing else requires that.

Also, we'd have to figure out for the first time where to put a new type like this that is custom to a specific operator.

[headinthebox]

100% agree, but I think it makes sense to adopt the Java8 overload.

[benjchristensen]

We can't as we compile to Java 6.

[benjchristensen]

We had to revert this because of ambiguity. See #1884 for the change.