Improve performance when apply bone transform in get_bone_global_pose()

[TokageItLab]

Improve performance when apply bone transform in get_bone_global_pose() by implementing bone_children and associating them with dirty flag and option: bypass updating skins.

The current implementation of Skeleton is that if you call get_bone_global_pose() after transforming a bone, recalculate transformation all the bones in the skeleton and access the visual (rendering) server for updating the skin. However, the current implementation causes a lot of waste waste when computing the bones in the same thread.

For example, a skeleton has 100 bones. If the transformation is applied to the bone at the base of the skeleton's right index finger, then the necessary recalculation would normally only require 3 times recalculation, so including the child. However, in the current implementation, it would be recalculated 100 times.

Also, under normal circumstances, applying the skin only needs to be done once at the end of the thread. If you want to transform a bone multiple times with get_bone_global_pose() in the same thread, slows down performance because accessing the visual server multiple times.

These issues have resulted in very poor performance when using get_bone_global_pose() in add-ons that simulate bone transformation, such as IK and DynamicBone. In order to lay the groundwork for future add-on development, I make two main changes in this PR

• Implements a list of bone children as a property, reducing unnecessary recalculation of bones
• Implements an option: don't update skins in get_bone_global_pose()

I actually saw a large performance improvement in the DynamicBone I created for 3D model's hair.

[TokageItLab]

I later found that the destination for improving performance is the same as #41322, although the way we define the dirty flags to reduce the number of calculations and how to bypass the visual server are different. However, I would like to implement bone_children and bones_root are useful in GDScript.

[TokageItLab]

Skeleton implementations have changed a lot during the IK implementation #51368. As discussed with @lyuma, it need to be refactored for improved performance in the future.