[FIX] Excel readers: Keep empty flags row of three-line header

[raashish1601]

Issue

Fixes #7276

Description of changes

ExcelReader.get_cells and XlsReader.get_cells dropped every empty row (filter(any, ...)) before the header was parsed. A file saved with type annotations but without class, metas or weights has an empty third (flags) header row. That row was dropped, so parse_headers saw the first data row in its place and did not recognize a three-line header. The names became a one-line header and the types row was read as data, which gives the strange table in the issue.

The readers now keep empty rows. ExcelReader still skips leading empty rows; XlsReader already starts at the first non-empty row. Empty data rows are still dropped, because adjust_data_width in data_table filters them, which is also how the tab/csv reader handles them.

Tests: test_write_file_without_flags writes iris without its class with with_annotations=True and reads it back. test_read_empty_flags reads new header_3_no_flags.xlsx/.xls files with an empty flags row through both readers. Both fail on master (the xlsx round trip gives 151 rows instead of 150, and the xls file reads a as a meta) and pass with the change. Orange.tests.test_io, test_tab_reader, test_txt_reader, test_xlsx_reader, test_table, Orange.data.tests.test_io and the distance matrix xlsx tests pass locally.

Includes

• Code changes
• Tests
• Documentation

[CLAassistant]

All committers have signed the CLA.