feat(array): introduce DataChunkBuilder and split chunks in HashJoinExecutor

[TennyZhuang]

The PR introduces DataChunkBuilder, which is responsible for two things:

1. Ensure that all columns are pushed with the same size.
2. Ensure that the size of the DataChunk should not be greater than capacity.

Signed-off-by: TennyZhuang zty0826@gmail.com

[wangrunji0408]

I have two suggestions:

1. Don't consume the builder when the size reaches the limit, so that users will no longer need to rebuild it by themselves.

impl DataChunkBuilder {
    #[must_use = "should yield the DataChunk when size limit is reached"]
    pub fn push_row(&mut self, row: impl IntoIterator<Item = DataValue>) -> Option<DataChunk>;

    /// Generate a [`DataChunk`] with the remaining rows, leaving the builder empty.
    #[must_use]
    pub fn take(&mut self) -> Option<DataChunk>;
}

2. Provide a helper macro to simplify the usage.

macro_rules! yield_if_some {
    ($maybe_chunk:expr) => {
        if let Some(chunk) = $maybe_chunk {
            yield chunk;
        }
    }
}

Then the usage should be something like:

let mut builder = DataChunkBuilder::new(...);
loop {
    // ...
    yield_if_some!(builder.push_row(values));
    // if necessary
    yield_if_some!(builder.take());
}
yield_if_some!(builder.take());

[TennyZhuang]

Don't consume the builder when the size reaches the limit, so that users will no longer need to rebuild it by themselves.

But it can force users to handle the finished DataChunk correctly by compiler error instead of just lint (

[TennyZhuang]

Provide a helper macro to simplify the usage.

Unfortunately, macros does not work well with yield.

[TennyZhuang]

@wangrunji0408 PTAL

[wangrunji0408]

LGTM in general. Sorry for the late response ==