In my experience, if we had an exact cache hit, the attempt to save the cache (under the same key) will produce an error.
So, the docs should be updated to something like

if: always() && !steps.<cache-restore-step>.outputs.cache-hit

at least in cases where we did a restore.

As cache-hit will only be true if the cache-primary-key equals the cache-matched-key this will always save the cache when it is possible.