This will require a new kind of Executor that executes both using the on-chain Wasm (WasmExecutor) and, when version allows as such, natively (using NativeExecutor).

Validator nodes should use both. Normal (heavy) nodes should default to native only when still syncing the chain, and then both when keeping in sync. A CLI option --use-native should allow strategy to be chosen (always (always use native, as long as version is good), never (never use native), syncing (as described here)).

Discrepancies between native and wasm should be loudly reported to the log and telemetry.