Currently, the client's writeConcern is passed on commitTransaction, since that is the behavior of TransactionRouter::commitTransaction.
TransactionRouter::commitTransaction stops processing responses and returns the last response on seeing a response with a writeConcern error.
However, unlike the ClusterCommitTransactionCmd, which just passes the return value of TransactionRouter::commitTransaction to the client, the change shard key code checks the response status and continues with the write batch. The code should probably also check the writeConcern status and report it in the correct format for writeConcern errors for batched writes.