ARROW-2226, ARROW-2233: [JS] Dictionary bugfixes

[trxcllnt]

Resolves https://issues.apache.org/jira/browse/ARROW-2226

[TheNeuralBit]

How does this bug present itself? I thought maybe DictionaryVector.slice wouldn't work, but I just wrote some unit tests for that on master and it seems ok

[wesm]

Would definitely be good to have a unit test for this

[trxcllnt]

@wesm@TheNeuralBit yeah, I wanted to get a fix in first. I encountered this trying to read values from a DictionaryVector whose indices have a nullCount > 0. I think this was just a typo/bad merge when I merged my branch with Brian's a few weeks ago. I would imagine the Dictionary integration tests could cover this case?

[trxcllnt]

@wesm I might've asked this before, but is it possible (via the spec) for the indices itself to be a DictionaryVector (e.g. for mapping between an original to sorted set of indices)? We don't guard against doing it on-the-fly, but if we want to support it, we should probably go through the public indicies.nullCount instead of using the private _nullCount here.

[wesm]

The indices are supposed to be a signed integer type

[TheNeuralBit]

I already started adding some basic unit tests for DictionaryVector, so I'll look at adding a test that covers the indices nullCount > 0 case and maybe PR them into your branch @trxcllnt

[TheNeuralBit]

Put up a PR with some tests: trxcllnt#5
I think I may have found another bug? Slicing a DictionaryVector with nullCount > 0 is throwing an error

[wesm]

@TheNeuralBit you should be able to push directly to this branch, too, if that's easier

[TheNeuralBit]

Oh nice, I'll do that in the future, thanks @wesm

[trxcllnt]

@TheNeuralBit lgtm thanks! 👍