RowCoverter::convert that targets an existing Rows

[alamb]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.
I am trying to make group by really fast in DataFusion apache/datafusion#6800

The grouping code uses the Arrow Row format 👍 and calls RowConverter::convert_columns

My traces imply that some non trivial amount of time is spent zeroing out newly allocated memory for Rows:

Describe the solution you'd like

I would like a method like RowConverter::convert_columns_in_place that writes to an pre-existing rows (clearing it out first)

Perhaps something like

impl RowConverter {
...
    /// Convert [`ArrayRef`] columns into pre-existing [`Rows`], first calling `Rows::reset()` 
    ///
    /// See [`Row`] for information on when [`Row`] can be compared
    ///
    /// # Panics
    ///
    /// Panics if the schema of `columns` does not match that provided to [`RowConverter::new`]
    pub fn convert_columns_in_place(&mut self, columns: &[ArrayRef], dst &mut Rows) -> Result<ArrowError> {

https://docs.rs/arrow-row/43.0.0/src/arrow_row/lib.rs.html#712

Describe alternatives you've considered

Additional context

[tustvold]

I'm not very familiar with the Instruments profile, but it seems off to me that alloc_zeroed is under update_group_state and not convert_columns? If the allocation was occurring as part of row conversion I would have perhaps expected it to appear under convert_columns? Is it possible there are other allocations taking place here?

[alamb]

I couldn't find anything else that does allocations per group, but I agree the profile is somewhat ambiguous

[alamb]

(BTW this is like 2% of the total time, so more like a nice to have feature, rather than anything critical or really important)

[tustvold]

label_issue.py automatically added labels {'arrow'} from #4533