Gives an alterative way to construct a Batch,
no good reason to keep them private.

Moves a point towards another point by an offset without overshooting.

These are needed often enough that adding them manually is a pain.

Implements airplane-like control of the pitch (elevation), yaw (bearing),
and roll (bank) angles of the camera. These angles are also known to as
Tait–Bryan angles.

* Color::clamp(): like in Vec and Point
* VecN::normalize_or_zero(): don't panic, return zero
* MatN::origin(): return the translation vector as a point
* Batch::append(): append the geometry of a batch to another
* Tri::centroid(): return the center point of a triangle

Supports linear, radial, and conical gradients with arbitrary color maps.

Implements the "slab method" described in https://en.wikipedia.org/wiki/Slab_method

The core module started getting really crowded.
Added a glob re-export to keep all existing paths valid.

Add proper unit tests for linear and translate/origin

* Implement fallback sqrt, finally
* Fix negative divisor in fallback rem_euclid
* Fix negative zero in fallback floor and rem_euclid
* Add exp and log2 to mm feature
* Improve mm sqrt and inv_sqrt precision
* Improve test coverage and consistency

* Triangle filling
* Triangle clipping
* Ray-bbox/sphere intersection

Revamp tests to ensure that the methods are approximate inverses.

* No square roots needed
* Division only when needed
* Early exits in unlikely cases only pessimize

* Add some #[inline]s and #[inline(never)]s
* Use unit-tagged matrix and vector types internally
* Avoid formatting machinery when not needed
* Polymorphize Lathe::build_with

Reduces generic bloat.

Can be used to reduce generic bloat and needless Vector wrapping.

Require that (a, b, c) passed to Plane::new is unit length.
Removes the need for normalizing on the fly and the abc() helper.

* Line-line
* Ray-line
* Ray-edge
* Edge-edge

These are all in the same "family", just different constraints.

…lized

This is a better match and allows making various Target impls generic over
any pixel format. The custom Framebuf type for SDL2 could be removed.

* Similar to SDL2 frontend
* Removes the unsafe block casting [u32] to [u8]
* Fixes component order from BGRA to RGBA, oops.