See discussion at https://discord.com/channels/273534239310479360/1215712247654125588

https://www.ralfj.de/blog/2022/04/11/provenance-exposed.html

Safe interfaces of `AnyVec` removed, as tested and usage constrained.

Inochi2D node types into Inox2D components, see comments in `node/components.rs`.
`Puppet` as self-managed struct, reads INP file into a tree of nodes and a world of components to construct self.
No `Copy/Clone` on `InoxNode`, components and `JsonValue`, cutting Aka.inp loading time to 1/3.

Storing absolute transforms is no longer RenderCtx's job.
Put graphics buffer stuff to a seperate file.

Should be easy for incorporating future renderable node types.

Get required data from components instead.
Changes to the renderer traits:
- remove `render()`
- `on_begin_mask()` -> `on_begin_masks()`
- `set_mask_mode()` -> `on_begin_mask()`
- `draw_mesh_self()` -> `draw_textured_mesh_content()`
- `draw_part()` -> `draw_drawable()`

…e backend gets smart.

For example reusing some results for a same node.

Enum `Deform` and `DeformSrc` to get prepared for Meshgroup implementation.
One `DeformStack` per deform-able node for combining multiple deforms in a controlled way.
Drawable related stuff, especially the drawable kind determination logic, moved to its own places from `render.rs`, as apparently other systems (parameters for example) may want to know is a node a drawable.

…oss frames.

so that `InoxNode` stores constant initial values.

The same reasoning behind `ctx` structs: Any data not in a model file is additional context data for certain functionalities, which should be only initialized upon demand.
End users only reset and set the values of params. When and how to actually apply params per frame is up to this library.
Removed `physics_ctx` placeholder that is present on local but should not be published.

General pattern for a "system": Takes the form of an optional `-Ctx` struct attached to a puppet, which can be initialized if dependencies met.
Interactions with the puppet data are done in methods of the struct, so a `None` system can do nothing.
Current systems: Transform, Param, Render.

With parameters and ECS system in place.
Changes to `trait InoxRenderer` during implementation:
- `new()` can have custom params.
- `on_begin_draw()` and `on_end_draw()`
Make `vao` creation, binding and unbinding, and vbo attaching, more explicit.
Move previous OpenGL `InoxRenderer` implementation into new framework.
Remove duplicate shader bind and uniform uploads during iteration over masks for a node (do once inside `on_begin_masks()`).

Properly implement zsort inheritance (+).
Correctly exclude composite children from `root_drawables_zsorted`.
Turn off param application temporarily, so this commit works for `Aka.inp`.

…d bug.

Currently, overflow will happen if value (0, 0) is applied on a param, which is clearly a bug.
The actual fix is not in the scope of ECS refactor.

The philosophy of the ECS refactor naturally leads to this cleanup:
In this refactor, spec-defined variables either goes into `InoxNode` or a component. If there are additional variables needed for the execution of a certain module, they are attached as additional components later.
It is revealed that bob position of a pendulum is such a variable.
After cleanup, all pendulum-like systems can provide methods for accessing and updating their bobs' position, implementing `trait Pendulum`.
And then `trait SimplePhysicsCtx` will be auto implemented for all `impl Pendulum`, handling the mapping from transforms to simulation input, from simulation input to parameter values.
More comments are added.

The implementation is inefficient (check changes in `puppet.rs`), yes, but clear, and the bottleneck is not here anyways.
Cloning of `SimplePhysics` is required for working around ownership problems, until dark magic implemented on `World` enabling multiple mutable handles to different components.

And suppress variant unused warning... until Meshgroups.

Responding to Inochi2D#86 (comment)
Rename `ParamMapMode` to `PhysicsParamMapMode` for this change for clarity.

In response to Inochi2D#86 (comment)

In response to Inochi2D#86 (comment)
Plus more robust `upload_array_to_gl()`, with more annotations inside.

In response to Inochi2D#86 (comment)
...and why `Sized` is required in the first place?

In response to Inochi2D#86 (comment)
The core crate does not need to know about the presence of stuff to do before and after `draw()`. This is up to renderer implementations.

…mbinations for a Drawable.

In response to Inochi2D#86 (comment)

…urce a `DeformStack`.

Thus accordingly, fewer `DeformStack` resetting and reading will happen.
Optimization inspired by Inochi2D#86 (comment)

For future meshgroup implementation.