Fixes a bug introduced by #105718

With allow_duplicates: false, we're pretty careful to not upcast a scalar into a list unless some kind of change has been made to the value -- that is, if you have a value of "2" and you are ignoring duplicates and you try to append "2" onto it, you still have a value of "2" and not ["2"].

With #105718, there's a new behavioral path, that is ignore_empty_values: true. We need to take the same care here that we don't upcast a scalar (or the absence of a value) into a list unless there's actually some value to be appended. That is, if you have ignore_empty_values: true and all the values that you want to append are actually empty, then the resulting operation should be a no-op, but it wasn't! This PR fixes that.

ALSO, I'm super pleased with how innerAppendValues and innerAppendValuesWithDuplicates drop out because of this, I've never been especially happy with those methods existing. 😄

Note: Given that there's another BC, I actually expect this to land in v9.2.0 (I imagine the labels will get corrected at some point).

Additional note: related to #104813, with was the feature request that asked for something like ignore_empty_values to begin with.