It looks like that the tdo_padoe_o signal is asserted too late at the begining of the shift_dr state.
Looks like it is not captured by the tests in adv_debug_sys because it is not used to tri-state TDO
top_io.v contains a simple fix : TDO is driven all the time