[Fix] Increase/decrease hash_key value for tokens depending on is_unlearn
When relearning a message, this script is called twice:
1. With `is_unlearn` true and the old `is_spam` value
2. With `is_unlearn` false and the new `is_spam` value
If `is_unlearn` is true, the code should not increase the value of the hash_key. Decreasing it ensures that it behaves like a fresh learn with the new is_spam value.
It seems there is no easy way to use int64 in Redis Lua, hence, we have
to use strings. It's much more expensive but still some advantage over
the previous schema.