fix(arrow-avro): decode records with no fields

[sqd]

An Avro record with no fields is legal, and encodes to zero bytes. The reader accepts such a schema and decodes its (empty) bytes fine, but fails at flush:

Invalid argument error: use StructArray::try_new_with_length or
StructArray::new_empty_fields to create a struct array with no
fields so that the length can be set correctly

Decoder::Record flushes through StructArray::try_new, which infers its length from the first child array, but a zero-field record has no child array.

RecordDecoder::flush (the top level) already handles this: it passes its own row_count through RecordBatchOptions::with_row_count. This change gives the nested case the same treatment. Decoder::Record now carries a row count, incremented wherever a row is appended and reset at flush.

Which issue does this PR close?

• Closes arrow-avro cannot flush records with no fields #10770.

Are these changes tested?

See included tests.

Are there any user-facing changes?

No

[Jefffrey]

perhaps we should name the fields, now that we have many fields

Record {
    fields: Fields,
    decoders: Vec<Decoder>,
    defaults: Vec<Option<AvroLiteral>>,
    projector: Option<Projector>,
    /// Useful when there are no fields (legal edge case)
    row_count: usize
}

[sqd]

thanks for the review! The fields are now named.

[Jefffrey]

thanks @sqd