It’s hard to say because the decentralized nature of things means that there isn’t somebody in charge who will define a timeline. The best I/we can do as devs is give the possibility for UIs to not use RPCs. And when it comes to that, the underlying tech is completely ready, it’s just on the frontend side that not everything is ready.
However it’s in principle completely possible for UIs to continue using RPC endpoints forever even if they have the possibility not to and even though RPCs are less secure. We can’t control that.
Full nodes will never remove the possibility to turn your node into an RPC node, because the RPC endpoint is how the owner of the node interacts with their node.
The only thing we want to deprecate is the fact that Alice uses an RPC node hosted by Bob even when Alice and Bob don’t know each other.
Is the plan to be able to also use it for historic data?
Light clients can’t provide old data at the moment. You still need RPC archive nodes for that. But it seems to me that most people use indexers (subscan and similar) rather than archive RPC nodes.
In the longer term (it requires BEEFY), light clients might be able to fetch historical data from archive non-RPC nodes. There are other problems to solve, so it’s a bit hypothetical.
To be clear, the full nodes that are currently RPC nodes are still necessary. It’s not possible to just “remove the servers”. Something has to provide the bandwidth and send data to the browsers.
The only change is that the browser would automagically find the full nodes and communicate with the full nodes in a way that doesn’t require a domain name in front of them.
So everything you said about treasury proposals and a collective is still relevant.