I assume that indices are raised/lowered with the flat metric at this stage? If that's correct, by far the easiest way is to re-declare your indices as 'type=free', e.g.
(use the same set as the original one of course). Then
canonicalise will do the job of collecting terms like you wrote here.