local entanglement entropy can be NaN

[jiangmi]

This is due to the MC process sometimes leads to negative
w = 1.0d0 - P0%up(i) - P0%dn(i) + P0%up(i) *P0%dn(i)
so that
ln(w) = NaN
This even cannot be avoided by calculating w in DQMC_Phy_Avg stage.