Future: differential outbox updates to reduce bandwidth beyond VALUE_PROP

[evnchn]

Tracking PR — do not merge

This tracks the broader bandwidth optimization idea from zauberzeug#5728: sending only changed attributes in outbox updates, not just skipping VALUE_PROP.

Why keep this

The bool-flag approach (zauberzeug#5933) only skips echoing VALUE_PROP back to the client. The original diff machinery from this branch could reduce bandwidth further by omitting all unchanged props, classes, styles, etc. For elements with large prop dicts (ag_grid with thousands of rows, plotly, table), this could be significant.

What blocked it

@falkoschindler's review identified:

• Performance risk: deepcopy on every element per outbox tick is expensive on hot paths
• Reconnection correctness: diffs in message_history could produce stale state on socket-only reconnect
• Falsy-key edge case: empty strings disappear from _to_dict, causing stale values on client

Path forward

If bandwidth becomes a bottleneck again:

1. Benchmark deepcopy vs json.loads(json.dumps()) on heavy elements
2. Consider copy-on-diff (only copy when something changed)
3. Ensure build_response and try_rewind always use full state
4. Handle falsy-key edge case in _to_dict

The original implementation is preserved on the outbox-simplification branch for reference.