add filter and filter! to GroupedDataFrame

[bkamins]

@nalimilan - this is a follow up to your comment in #1732.

[nalimilan]

Do we have any other in-place functions on GroupedDataFrame already, apart from select! and transform!? I'm asking because for consistency with the two latter, it would make sense for filter! to drop rows from parent(gdf). Though that could be a unexpected. We need a general policy on this, which could also apply to in-place oeprations on SubDataFrame.

[bkamins]

This is a good question (actually I initially considered to add filter only for these reasons). It applies both to filter and filter!. I have not thought about it because what I proposed was most natural.

So the considerations are:

1. we do not have other functions that mutate GroupedDataFrame in-place
2. we do not have functions that mutate SubDataFrame schema in-place (of course we allow mutating the data, but not schema)
3. select! and transform! work only for some GroupedDataFrames (ones that have no groups dropped) and they guarantee to retain row count in the parent; also it is in general expected (and documented) that they mutate the parent as they define new columns in general
4. filter! and filter can work on an already "subsetted" GroupDataFrame so then it would be confusing what should happen with groups that are already not present.
5. it is natural to expect that filter(predicate, collecion) does the same as collection[predicate.(collection)], which does not mutate the parent of collection if it is present.

All in all I think it is better to keep what we have.

[bkamins]

I have also pushed updates to docstrings of filter/filter! for AbstractDataFrame to make them consistent.

[bkamins]

Your comment made me realize that in the past we assumed GroupedDataFrame is immutable, and we used this assumption in transformation code. I have pushed a fix.

[bkamins]

This should be rebased after #2280 is finished and only then merged.

[bkamins]

So the only downside of filter! is that it modifies a view. I have reviewed all the code and in #2280 I make sure that we do not make assumptions that it is not modified, but still maybe it is better not to add filter!?

[nalimilan]

Maybe let's leave it out until we have a clear use case for it? If nobody requests it maybe it's not needed, and that keeps the API more consistent regarding what ! does.

[bkamins]

OK - I have removed filter!. This PR is independent now of #2280 as a consequence.

[bkamins]

similar - only coverage fails

[bkamins]

Thank you!