A couple of notes:

1. You should ideally make an accompanying proposal when adding new stuff, even if it's just exposing existing data like this, so as to motivate why you need it. You mentioned why you want the point count, but why you need the AABB was less obvious. We need to make sure we're solving whatever root problem you might be having.
2. PhysicsServer3D.soft_body_get_point_count does not need to be exposed. It will never be any different from the size of the index buffer of the mesh you've assigned to your SoftBody3D. See the discussion in godotengine/godot-proposals#13335 for more details. There might still be some argument for keeping SoftBody3D.get_point_count() though, to simply return a cached value of the index buffer size.
3. Since SoftBody3D inherits from VisualInstance3D, it already has a virtual get_aabb() method that should be used instead of adding a new get_bounds() method. This could in fact be framed as a bug even, since the editor won't currently show the correct bounds when you select a SoftBody3D using the "3D" input mode in the embedded game view, but will if you implement get_aabb() instead.

(Also, calling the physics server getters directly, like what so many of the methods in SoftBody3D do, is quite problematic for threading reasons as well, but that's beyond the scope of this PR I guess.)

Since it is apparently easy to overlook that you can query the mesh for the number of vertices in the soft body, an obvious function that gives you the number seems useful to me.

Seems to work well, and looks good to me.