Fix reading multiple complex data types exception

[harveyyue]

Fix #139

[Jefffrey]

Thanks for raising this. CI failures look unrelated, should be fixed by #141 so just need to pull from main.

Could you also include a short explanation for this fix, to help me understand how it fixes the issue?

[Jefffrey]

Suggested change

	let mut projection = Vec::with_capacity(projected_schema.fields().len());
	for named_column in builder.file_metadata().root_data_type().children() {
	if let Some((_table_idx, _table_field)) =
	projected_schema.fields().find(named_column.name())
	{
	projection.push(named_column.data_type().column_index());
	}
	}
	let projection = builder
	.file_metadata()
	.root_data_type()
	.children()
	.iter()
	.filter(|named_column| {
	projected_schema
	.fields()
	.find(named_column.name())
	.is_some()
	})
	.map(|named_column| named_column.data_type().column_index());

Thoughts on using a more iterator based approach? Could potentially be simplified further.

[harveyyue]

Could you also include a short explanation for this fix, to help me understand how it fixes the issue?

The elements from complex data type like map will take up indexes as below

[Jefffrey]

Could you also include a short explanation for this fix, to help me understand how it fixes the issue?

The elements from complex data type like map will take up indexes as below

But how does this relate to the code changes you made?

[harveyyue]

Could you also include a short explanation for this fix, to help me understand how it fixes the issue?

The elements from complex data type like map will take up indexes as below

But how does this relate to the code changes you made?

We need to use origin projection to get arrow projected schema, and do the mapping to get orc NamedColumn from this arrow field name, will get the real column index.