SCALAR_TABLE_FUNCTION: add mapping vector

[Tishj]

		// Any other vector type: we need to create a selection vector
		// but can avoid flattening because the memory of the selection vector will not be temporary
		AddProjectedColumnsFromOtherMapping(base_columns, input, intermediate_chunk, chunk);
		break;


void PhysicalTableInOutFunction::AddProjectedColumnsFromOtherMapping(idx_t map_idx, DataChunk &input,
                                                                     DataChunk &intermediate, DataChunk &out) const {
	auto &mapping_column = intermediate.data[map_idx];

	SelectionVector sel_vec;
	sel_vec.Initialize(intermediate.size());
	// Create a selection vector that maps from output row -> input row
	UnifiedVectorFormat mapping_vector_data;
	mapping_column.ToUnifiedFormat(intermediate.size(), mapping_vector_data);
	auto mapping_data = reinterpret_cast<sel_t *>(mapping_vector_data.data);
	for (idx_t i = 0; i < intermediate.size(); i++) {
		// The index in the input column that produced this output tuple
		idx_t idx = mapping_vector_data.sel->get_index(i);
		const auto input_row = mapping_data[idx];
		D_ASSERT(input_row < STANDARD_VECTOR_SIZE);
		sel_vec.set_index(i, input_row);
	}

	// Add the projected columns, and apply the selection vector
	for (idx_t project_idx = 0; project_idx < projected_input.size(); project_idx++) {
		auto source_idx = projected_input[project_idx];
		D_ASSERT(source_idx < input.data.size());
		auto target_idx = map_idx + project_idx;

		auto &target_column = out.data[target_idx];
		auto &source_column = input.data[source_idx];

		target_column.Slice(source_column, sel_vec, intermediate.size());
		// Note: we can avoid flattening this
	}
}