Added ScaleRow to GtkHelper - #34

… settings access

Three related ordering defects (issue StreamController#34):

- on_ready_called is set at schedule time (it must be: plugin API calls
  inside on_ready gate on it via raise_error_if_not_ready), but
  own_actions_tick treated it as 'ready' -- on_tick ran before on_ready
  had even started. New ActionCore.on_ready_finished is set in a finally
  after on_ready returns/raises, and ticks gate on it.

- own_actions_update dispatched on_update mid-initialization; the default
  on_update calls on_ready (compat), producing a second on_ready running
  concurrently with the pool's initial one (duplicate backend processes).
  Updates now skip until on_ready_finished; nothing is lost -- the initial
  ready sequence ends with its own on_update.

- Actions of one plugin run on_ready in parallel since the pool-based page
  load, so get_settings' v1->v2 conversion write raced concurrent
  set_settings: a stale conversion could clobber a fresh save. A per-plugin
  RLock (lazily created for __new__-built instances) serializes the
  accessors. set_settings' read is also hardened against a corrupt existing
  file -- the same class get_settings already guards.

The new scenario red-checks all three against the unfixed code: 8 ticks
delivered pre-ready, 2 concurrent on_ready entries, and a v1 conversion
clobbering a concurrent save.

…treamController#16, StreamController#127)

- Page.set_media_fps live-applied to matching inputs on ALL controllers:
  editing one page's FPS row rebased the playing video timeline on another
  deck showing a different page (self-corrects only on that deck's next
  page load). Now filtered by active page, the same filter update_input
  uses (issue StreamController#14).

- mark_page_ready_to_clear dereferenced active_page at call time; both
  bracketing callers (the tick loop and the key handler) re-derefed it for
  the True-call, so a page switch during a slow on_tick left the OLD page
  pinned ready_to_clear=False forever -- unevictable, silently shrinking
  the eviction budget. The False-call now returns the page it marked and
  the True-calls pass it back (issue StreamController#16).

- Page.initialize_actions' bare check-then-set on on_ready_called let two
  concurrent load_page(samePage) calls both claim the same action and
  submit a second concurrent on_ready (duplicate backend processes -- the
  class StreamController#34 closed for the on_update path). The claim is now atomic under
  a per-page lock; the loads and pool submit stay outside it (issue StreamController#127).

tests/scenario_small_guards.py red-checks all three: the FPS bleed, the
pinned page, and (via a capture-before-rendezvous barrier property making
the interleave deterministic) the double claim.