Missing bit from #10374: because the _read_xls function is a generator
it is *entirely* lazy and Python doesn't execute *any* code when it's
initially called, which means the file is actually (tentatively) parsed
when the first line is requested which happens outside of
_read_file (and thus of the fallback mechanism implemented by #10374).
Fix by splitting _read_xls between the opening of the workbook (eager)
and the actual iteration (which remains lazy).
Xavier Morel
4a31f85c5a