Geth LES rewrite (chance to get rid of centralized RPC 4 large number of dapps)

The author of this geth forum post proposes a full rewrite of the Light Ethereum Subprotocol (LES), arguing that the current design, which they say the merge essentially killed, tried to do too much. They propose that light servers provide only a strict subset of full-node data, limited to transaction indices at most T time old, with a couple of weeks or the last 128 blocks suggested as options. They also argue for replacing LES's complex flow control with simple token buckets, dropping devp2p at this layer in favor of an HTTP API server, and adding small friction points, such as serving HEAD-1 or HEAD-3 instead of the latest block and a slight response delay, to discourage abuse.