Feat/deterministic metadata encoding

[timsaucer]

Which issue does this PR close?

None, but I can open one if necessary.

Rationale for this change

The ordering of metadata is not consistent since it uses a HashMap. It can be useful in unit tests to verify an output from a known hash of it's serialized values. With metadata this is not consistent.

What changes are included in this PR?

Adds ordering to the hashmap keys when encoding.

Are there any user-facing changes?

No.

Example

If you run this example multiple times, you will see the encoding changes from run to run based on the non-deterministic ordering of the hashmap iterator.

use std::{hash::Hasher, sync::Arc};

use arrow::{array::RecordBatch, datatypes::Schema};

fn main() {
    let schema = Arc::new(
        Schema::empty().with_metadata(
            [
                ("a".to_owned(), "1".to_owned()), //
                ("b".to_owned(), "2".to_owned()), //
                ("c".to_owned(), "3".to_owned()), //
                ("d".to_owned(), "4".to_owned()), //
                ("e".to_owned(), "5".to_owned()), //
            ]
            .into_iter()
            .collect(),
        ),
    );
    let batch = RecordBatch::new_empty(schema.clone());

    dbg!(&batch.schema().metadata().keys());

    let mut bytes = Vec::new();
    let mut w = arrow::ipc::writer::StreamWriter::try_new(&mut bytes, &schema).unwrap();
    w.write(&batch).unwrap();
    w.finish().unwrap();

    let mut h = std::hash::DefaultHasher::new();
    h.write(&bytes);
    let h = h.finish();

    eprintln!("{} bytes -- h = {h:x}", bytes.len());
}

[timsaucer]

I currently based this off 54.3.1 but I will updated it to main after we have completed internal testing.

[timsaucer]

Rebased on main, ready for review.