Reintroduce BeforeBake* and AfterBake* model modifiers

[PepperCode1]

This pull request effectively reverts 2424258 and adds some minor improvements, including a fix for #4310 and documentation improvements.

The decision to reintroduce these events was made due to the following reasons.

• Wrapping unbaked models could lead to duplicate baked model wrapping. An unbaked model with no elements and a non-null parent delegates baking to the parent, so if both the child model and parent model are wrapped, the baked model produced by the parent will be double wrapped, even though that's usually not preferable. This is not too much of an issue for simple wrappers in an isolated environment, but in the general case with multiple mods doing arbitrary wrapping, correctly handling duplicate wrapping becomes complex or impossible.
• Mods may need to wrap certain models based on conditions that cannot be computed as early as when the OnLoad* events run. While such mods can still use the OnLoad* events to accomplish their goal, it requires wrapping the unbaked model unconditionally and doing the real wrapping within UnbakedModel#bake or GroupableModel#bake. This pollutes the vanilla code path with model wrappers even when it's not necessary and can increase the amount of allocations.
  Additionally, such mods cannot benefit from two of the three main reasons previously used as the basis for the removal of the *Bake* events (those being the one related to model resolution and the one related to block model groups).
  One specific mod for this reason is Continuity, which only wraps models based on which sprites are present in the atlas. Model loading and atlas loading/stitching happen concurrently, but model baking happens after both.