Ensure GroupedDataFrame starts at index 1

[nalimilan]

Storing indices of a ghost group could be confusing and makes the code more complex later.

[bkamins]

Looks good and copyto! should work for overlapping blocks of data looking at the code, although it is not a part of the guarantee given in the documentation. Maybe we should ask in Base to confirm and put into the documentation the contract that overlapping blocks of memory may be passed to copyto!.

[bkamins]

have you considered using a view instead of copyto!? I think it should be faster and use less memory.

[bkamins]

We might also consider avoiding considering missing group starting from:

rperm = Vector{Int}(undef, length(groups))

• make this vector shorter from the start and drop first starts and stops if needed. This should be even a bit faster when there are missing values as we would do less copying.

[bkamins]

As a second thought - it would probably slow down things a bit when we have a lot of small groups.

[bkamins]

If you are happy with current #1689 I think you can close this PR (unless you want to clean it up to always start from 1, but as discussed it is probably not strictly necessary)