A state var named _get_was_touched silently disables disk-state persistence

[masenf]

• Cluster: event_hotpath | Regression: no | Verifier: pending
• Repro: add _get_was_touched: int = 7 to a state, run with the default disk state manager, send one
  event, wait past the flush debounce: every flush logs Error processing write queue: TypeError("'int' object is not callable") and shutdown ends in a traceback from
  istate/manager/token.py get_and_reset_touched_state. State is never persisted. Reads and writes of
  the var in the browser work, so the failure is invisible from the UI.
• Note: perf: cut per-event overhead on the state/event hot path (−28% CPU/event, +20% throughput) #7025's own unit test uses this exact name as a supported collision, while the runtime call
  site (which predates the PR) calls it as a method.
• Evidence: event_hotpath/logs/gwt_disk_smoke.trimmed.log, gwt_disk_base0910.trimmed.log.

Expected: reject the reserved-name collision at state definition time or preserve the internal method call so disk persistence succeeds. Related #7074 / #7077 concern inherited base-var shadowing; this is a collision with a framework method, so those reports do not cover this failure.

Source: prerelease finding 010, recorded against Reflex 0.9.11a1 and compared with 0.9.10.post2. See the full campaign report and cluster notes and reproduction artifacts. Findings and verification status above are from that campaign; this filing does not claim a new browser rerun.

[linear-code]

ENG-12355