I think that you are describing “segregated witness”, which was originally from Bitcoin, but it has been brought up in Eth1x discussions. Also relevant are access lists, which are gaining support, and are a step towards segregated witness.
BTW some things you mention have changed since the Eth1x pivot to ReGenesis. For example, under ReGenesis, people want (i) transaction witnesses (which seem to solve the witness gas accounting problem), (ii) a KV-witness encoding (which is more convenient for ReGenesis), and (iii) no code merkleization (which is awkward and less critical under ReGenesis). This is just the feeling I get from following Eth1x discussions.
If one thinks more abstractly, then they may want to decouple witnesses from blocks entirely. One hopeful idea is to use pre-consensus, an avalanche-like consensus to agree on which witnesses are important before blocks are mined. Another discussion was about having a tendermint-like multi-stage consensus with delayed execution to allow witnesses to propagate.
Kate commitments are not currently being championed for Eth1x. Hashing is less controversial and more practical for now. We hope that cryptographic accumulators will mature in a few years. Or maybe the accumulator problem will disappear.