Fixes #112566.

Ensures SpriteBase3D::parent_sprite is set on NOTIFICATION_PARENTED/NOTIFICATION_UNPARENTED instead of on NOTIFICATION_ENTER_TREE/NOTIFICATION_EXIT_TREE. This was already enough to fix #112566 as the bug happened because SpriteBase3D::_draw() was called before the sprites entered the tree:

• the redraw was queued when outside the tree, during scene instantiation when setting stored color/texture,
• the MessageQueue was flushed before entering the tree, during physics processing (it's irrelevant where exactly),
• the child sprite's dirty-marked color was updated as if it had no parent, and was marked as not dirty anymore with such incorrect cumulated-color,
• scene was changed "at the end of a frame" (scene changes are deferred on their own, independent of MessageQueue/call_deffered calls).

Moreover, after the above there was another redraw triggered, as on NOTIFICATION_ENTER_TREE it would update/redraw immediately in case no redraw is enqueued, which was the case when reloading the scene (the another redraw would use wrong cumulated-color for the child sprite).

Redrawing being triggered when outside the tree seems pointless to me. 🤔
Hence I've made it not update when outside the tree, and update immediately on entering the tree. This makes e.g. on reloading the scene as in the MRP from #112566 the SpriteBase3D::_draw() be called only once per Sprite3D (instead of twice, including the first one outside the tree).