Forest loss driver prediction pipeline gets dates wrong

Some of the forest loss events have date set to 2019/2020 even though they should all in the last year.

Actually I think all the errors are using date 2019-01-01 (which is the "zero date" in the GLAD data since it measures dates as days since 2019-01-01).

Anyway somehow the date is taken from a point outside the actual forest loss event polygon.