Corrected dual objective returned by PDLP

[Yakun1125]

I read that in Version 9.8 and later, the dual objective is reported as replacing the infinite terms in the objective value with finite ones.
But I don't quite understand why it does not necessarily give a lower bound on the objective value?

The LP I am solving always has the lower bounds of all variables are set to zero and the upper bounds are set to infinity. But the dual objective value reported by the PDLP algorithm does not consistently provide a lower bound on the primal objective value. In some cases, the dual objective value can be larger than the primal objective value.

Are there any suggestions to obtain a safe lower bound on the objective value?