Questions about index module to solve performance issues due to limitations of modular architecture.

[khanh-to-niteco]

Hi Medusa team,

Please help to elaborate more on the index module to solve this performance issue: #13356
As confirmed by @olivermrbl

1. Can we create custom indexes on the indexed data since this is a very common use case? How does it work? It would be helpful to add more technical details like this in Medusa docs.
2. The indexed data looks like a "pre-join" data between tables. In the docs, i see examples of building indexes for 2 relations, what if we need to join more than 2 relations? Since this is a pre-join table, it seems the data could grow very big if we join more than 2 relations. It would be helpful to add this into the docs, too.
3. Looks like the index still does not solve the issue since it still gets the list of matching IDs to pass to the main datastore: [Bug]: Performance degradation in maybeApplyLinkFilter - fetches all related IDs causing 10x slower response times #13356 (comment)
4. If we de-normalize the data to store in the index so we don't need to query the data store, we may be able to solve 3, but it raises other questions:
   a. In many-to-many or one-to-many relations, we could have a lot of duplicate data. When we update 1 item in the main datastore, the index has to update many duplicated rows (let's say a few dozen thousands in our real use case)
   b. Would issue 2 above makes this even worse?