Jaxrenderer

1. Change the ordering of quaternions (in `geometry.py`) to `(w, x, y, z)` instead of `(x, y, z, w)` to be consistent with the convention used in `pytinyrenderer` and `BRAX`. Reference: [brax/math.py](https://github.com/google/brax/blob/aebd8b8cb34430f6eaf6f914293f901e3c8d9a22/brax/math.py).
2. Fix: remove unnecessary `staticmethod` decorator in `merge_objects`.
3. Changed the way that `Camera` is created in `Renderer.create_camera_from_parameters` to force convert parameters into `float` weak type.
4. Force convert `LightParameters` to JAX arrays in `Renderer.get_camera_image` to avoid downstream errors.
5. Downgrade minimum Python version to `3.9`, `numpy` version to `1.22.0`, `jax` and `jaxlib` version to `0.4.4`.

1. Change the default behaviour of `renderer/utils.py::transpose_for_display` which will flip vertically as well by default, so the origin of the resultant matrix will be (height, width, channels) and with the origin located at the top-left corner. The previous behaviour can be achieved by setting `flip_vertical=False`.
2. `Scene.add_cube` now accepts one number for `texture_scaling` to scale texture map equally in both x and y directions.
3. Fix some assert message issues (in `Scene.add_cube`).
4. `CameraParameters` now accepts `position`, `target` and `up` in Python's tuples of floats as well, along with `jnp.array`.
5. `Scene.set_object_orientation` and `Scene.set_object_local_scaling` supports tuple of floats as well as inputs, additional to `jnp.array`.
6. `Model` now has a convenient method `create` to create a Model with same face indices shared by `faces`, `faces_norm` and `faces_uv`, and a default `specular_map`. This is useful for creating a mesh where all vertices has its own normal and uv coordinate specified, under same order (thus same face indices).
7. Correctly support Python Sequence for `utils.build_texture_from_PyTinyrenderer` as texture.
8. `quaternion` function to create an orientation from axis and angle, and `quaternion_mul` to composite quaternion.
9. `rotation_matrix` function to create a rotation matrix from axis and angle. Also allows `Scene` to set object orientation directly using rotation matrix.
10. Move `Renderer.merge_objects` into `geometry.py`, and expose in `__init__.py`.
11. `batch_models` and `Renderer.create_buffers` convenient method to facilitate batch rendering of multiple models.

1. For built-in fragment shader outputs, a new field `use_default_depth` is added (default `False`) instead of relying on checking if `gl_FragDepth` is `NaN`.
2. Change `ZBuffer` value meaning to OpenGL convention: all values are non-negative, and the smaller the value, the closer the point is to the camera.
3. Change default `Shader.mixer` behaviour: now it picks the fragment with minimum depth value `gl_FragDepth`.
4. Change `shadow.py::Shadow.render_shadow_map` according to the z-value definition change, to add into the shadow map instead of subtracting so to move shadow maps further away from the "camera" (light).
5. Change `shaders/phong_reflection_shadow` according to the z-value definition change.
6. Transform shadow coordinates in VS (world => NDC) and FS (NDC => screen) instead of transforming in FS (model's screen => world => screen) to avoid precision loss in inverse matrix computation.
7. Fix issue of perspective-correction barycentric interpolation in `pipeline`.
8. Fix `shaders/phong*` so normals are correctly transformed into pre-projection eye coordinates, rather than being projected.
9. Rename `ModelView` => `View`, `model_view` => `view`, `model_view_matrix` => `view_matrix` as the matrix is actually view matrix that transforms from world to eye space, not model view matrix (model to eye space).