Rapid Stage Select hardened against watcher loss (T015, addon v0_91)
The Rapid Stage Select watcher (F022, D119) could stop firing after an undo/redo, a Navigator visit or Reset GAAD Platform, because Blender clears bpy.msgbus subscriptions across those events and a breed interrupted before its finally ran could leave the re-entry busy flag latched. v0_91 makes the watcher self-healing through one shared recovery helper, _gaad_rapid_rearm(), which clears the busy flag and re-subscribes the active-object watcher. It is now called from four places: new undo_post and redo_post persistent handlers (the prime cause), at the end of the Reset GAAD Platform operator, on file load (replacing the load-only re-arm of v0_79), and from a new update callback on the gaad_rapid_select toggle so flipping it OFF then ON is a user-accessible recovery. Separately, the busy flag now carries a time() stamp and self-clears after a 2 s timeout in the notify guard, so an interrupted fire path can never wedge Rapid permanently. Re-subscribing does not itself breed: subscribe_rna notifies only on later changes, so the active object restored by an undo is not treated as a click. The change is additive - while Rapid is working normally there is no behaviour change - and it does not touch the single gaad.generate breeding path (D119).