Runtime & worker: optimistations for large state objects

I want to think about how to optimise large state objects

At the end of each step we do a bunch of processing on objects, including:

• cleaning up the returned state
• serializing the state object out of the worker thread
• serializing the state object out of the child process
• serializing the state object through the websocket
• trying to work out the size of the object (once or many times)
• I think there's some stuff about removing circular references

I'd like to explore things like:

• using shared memory buffers to reduce serialization
• using heuristics to guess that state object is large (Ie, an array of 10k items, or an object more than 100 properties deep), and then optimising around that (like bailing out of serialization and circularity checking earlier)

Start at the end-of-step processing for state objects in the runtime and worker, tracing cleanup, size checks, circular-reference handling, and serialization through the worker thread, child process, and websocket. Compare the costs of these paths for large state objects and define benchmarks and acceptance criteria before choosing an optimization; the issue does not name files or tests.