Prefilter translation before join
When searching on a translated field, we have to make a join with the ir.translation
table but we could apply the filter on the value
on the left part of the join. This will reduce the number of tuple to make the join.
The only problem is that we can no more guarantee to use the last duplicated translation like #10061 (closed) fixed. I think this is minor compare against the performance gain. Also #10061 (closed) update all the duplicated translations so their value should be the same and it is abnormal to have duplicated translations.