Right now my current understanding is that a basic 3D model animation system needs three things:

1. A way to load keyframe+pose+bone data from somewhere (Some engines have it in-engine, Bevy supports loading from GLTF/GLB which means it'll load your Blender animations and rigs along with your models) and play back a selected animation with basic controls like repeat/pause/etc. In Bevy this is the AnimationPlayer and there's an AnimationClip and it's all tied in to the Gltf class too
2. A way to combine the individual animation clips from 1 into a graph structure to allow for blending of multiple animations or masking of individual bones (this allows you to, for instance, attack and walk at the same time and play both animations and either blend them together or only have individual bones from each move). Bevy has AnimationGraph for this (and in fact even if you don't want to use the blending/masking abilities you still need to make an AnimationGraph from a single clip) which you attach to an AnimationPlayer using AnimationGraphHandle and then the player can control which animations are playing.
3. And a third thing - a state machine of some kind to determine which animations should be playing. Are we idling? Swinging a sword? Getting hit? It all came together when I watched a Unity tutorial and I saw the graph structure in the animation system there - basically it can respond to external triggers and also internal logic to shift between states, which have one or more animations playing. Bevy doesn't currently have something tailor built for this so I'm uh stumbling through making my own.

There are seams and dumb assumptions I made everywhere and I don't have a working demo yet but the code is starting to come together, its just a lot of balancing simplicity vs flexibility for the project I'm working on since I don't plan to re-use this code.

Of course, animating UI and 2D is a whole other thing.