While working on the refactoring PR #1933, I noticed that there is no _gej clearing of the point multiplication result in _ecmult_const_xonly (apparently this was missed back then when looking for instances in #1579 (comment), likely because in this case there is no _ge_set_gej call involved).
Note that for _ecmult_const itself there is already a separate PR: https://github.com/bitcoin-core/secp256k1/pull/1885