PAM Native 1.30.1

• Commits between frames: a batch that builds something (32 mutations or
  more: a page of older chat rows, a reel's chrome) and any batch while a
  virtual list is dragged or flinging is committed right after the running
  frame instead of inside its animation callback, then the virtual lists it
  dirtied are measured and laid out there too, at their current size. The
  frame keeps only what was already running (the scroll, a playing video);
  the work uses the idle time before the next vsync while the render thread
  draws, and the next traversal shows the rows. Small batches (a keystroke
  echo, a toggle) still land in the frame that asked for them. iOS defers the
  same heavy batches to a one-shot beforeWaiting observer ordered after
  Core Animation's commit.
• Android ActivityIndicator: the spinner is a plain View drawing the same
  Vuetify-style arc instead of a ProgressBar, whose constructor loaded the
  platform's indeterminate AnimatedVectorDrawable even without a style
  (15-25 ms on the UI thread for every new spinner, e.g. a list's "loading
  older rows" cell, on Samsung). Accessibility still reports a ProgressBar.
• Android text cells: a pooled TextView taken for a new list cell no longer
  keeps the previous cell's size, so setText defers to the single
  measure instead of laying the text out at the old width and again at the
  new one.
• Runtime statistics (RuntimeFrameMetrics.stats) are read from the engine on
  first access (the dev tools) instead of on every commit, where the JNI call
  cost up to milliseconds on the UI thread. Benchmark builds log
  batches/decodeNs/mountNs per commit without them; debug builds keep the
  full line. iOS reads them lazily too.