I ran into an identical problem, except pandas was oddly converting only some cells into datetimes. I ended up manually converting each cell into a string like so:
def undate(x):
if pd.isnull(x):
return x
try:
return x.strftime('%d/%m/%Y')
except AttributeError:
return x
except Exception:
raise
for i in list_of_possible_date_columns:
df[i] = df[i].apply(undate)