Add Fetch API to typed fields

[xichen2020]

cc @notbdu

This PR adds Fetch to typed field interfaces, which allows us to fetch values from typed fields for a given doc ID set determined from a doc ID set iterator passed in.

[black-adder]

do you see value wrapping the docIt inside a wrapper type that implements SeekableDocIDSetIterator? ie the wrapper implements SeekForward and uses this code here. Inside NewAtPositionDocIDSetIterator, you always wrap docIt inside a SeekableDocIDSetIterator if it doesn't already implement it. Should make this function more readable.

[xichen2020]

Yeah possibly, although a bit wary of introducing yet another wrapper type that can be implemented in a few lines. I'd say if there's another place that needs this logic then yes, otherwise maybe not.

[black-adder]

nit: s/newbi/newBi

[xichen2020]

Sounds good.

[black-adder]

isn't this more Intersect than Fetch? also, isn't this more appropriate given this is supposed to be a "set"

[xichen2020]

Yeah my thinking was if at some point we want to treat the doc IDs that are in the set passed in but not in the current set differently then the name Fetch could potentially make more sense, but I can change it to Intersect as well.

[black-adder]

treat the doc IDs that are in the set passed in but not in the current set differently

Maybe we can call that case difference? I do like the idea of using traditional set function names.

[xichen2020]

This is semantically Intersect though, change included in the next PR

[black-adder]

I might be totally misunderstanding something; I was under the assumption that doc ids are monotonically increasing, ie the DocIDSetIterator might start at 1000 and return 1000, 1001, 1002, etc. However here we're comparing against a full doc ID set representing [0, numTotalDocs) which can only occur naturally once given the doc ids are monotonically increasing.

[xichen2020]

Hm not sure I get what you meant by "can only occur naturally once"?

[black-adder]

given it.currDocID < it.numTotalDocs is an invariant, can we simplify this to:

         if it.done {
		return false
	}
	if !it.maskingIt.Next() {
		it.done = true
		return false
	}
	it.currPosition++
	it.currDocID = it.maskingIt.DocID()
        return true

[xichen2020]

Would prefer to be a bit defensive here in case the masking iterator goes out of range.

[black-adder]

can this not be done upfront in the constructor so we don't have to check it every time?

[xichen2020]

Yeah sounds good, can do.