In a sense, I believe this is already been done from the direction of the
mongos pulling the change instead of the config servers pushing the change.
For metadata change (chunk split, chunk move, etc.), the mongos will load the metadata lazily. For some changes, such as config server member change, the monitoring thread on the mongos will ping periodically and notices the change. Are these the cases you have in mind?
However, some major change e.g. movePrimary would require all
mongos routers to be restarted in pre-MongoDB 4.4.