Restructure Index trait to allow for more extensive Index struct implementation.

[y-pakorn]

This should not affect/or break anything. Trait change allows for the implemented Index struct can access both old_data and data and act accordingly.

For the ideas, In some cases, we will not need to update the indexes every update, only handle the first save and remove. This can be called ConstantMultiIndexes.

Or we can update the indexes whether the condition of the data updated is met. This can be called ConditionalMultiIndexes.

pub trait Index<T>
where
    T: Serialize + DeserializeOwned + Clone,
{
    fn update(
        &self,
        store: &mut dyn Storage,
        pk: &[u8],
        old_data: Option<&T>,
        data: Option<&T>,
    ) -> StdResult<()>;

    fn save(&self, store: &mut dyn Storage, pk: &[u8], data: &T) -> StdResult<()> {
        self.update(store, pk, None, Some(data))
    }

    fn remove(&self, store: &mut dyn Storage, pk: &[u8], old_data: &T) -> StdResult<()> {
        self.update(store, pk, Some(old_data), None)
    }
}

[maurolacy]

For what I can see in the associated PR, this is just cosmetics: Moving replace's functionality inside of update, and handling both save and remove in terms of update.

I wouldn't merge this unless there's a good reason that indicates this is better than the current implementation.