feat(studio): optimistic in-memory edits (near-instant canvas) + content-first inspector

[LeadcodeDev]

User iteration round 4:

1. Near-instant feedback: edits currently reach the canvas in ~300-550ms (250ms write debounce → notify watcher → 250ms hot-reload poll → render). Requested quasi-instantaneity. Frozen design: optimistic in-memory apply — every edit immediately mutates the in-memory raw + rebuilds the scenario (JSON re-resolve / HTML retranspile) + bumps generation + nudges the rev signal directly (no poll wait); the canvas refetches and the prefetcher re-primes within one render (~15-25ms release). The 250ms debounced disk write continues in background (undo/history unchanged — still keyed on effective disk writes); the watcher skips reloads whose content hash matches the last self-write (no redundant reload/flicker, external edits still reload).
2. Content first: for text-family components the CONTENT section renders above PROPERTIES.