SetUserData is no longer needed since a non-const reference is provided by GetUserData.

Switch from travis to Github Actions for CI and testing

Small fixes for: #681, #686, #689, #690, #691, #692, #694, #695, #696
Fix workflow for main

nitially I tried to make gift wrapping handle collinear points better. Conclusion: gift wrapping is a bad algorithm.
So I removed gift wrapping and implemented quickhull. I also added code to explicitly remove collinear points. This turns out to be quite easy once you have a convex hull.

I ran over 200 million randomized tests successfully with test data that is very likely to have collinear points with small numerical deviations. All the tests passed and algorithm correctly rejects very close points and fully collinear points.

I also tested performance at around 150 milliseconds per million hulls.

I split out the convex hull code into a separate function b2ComputeHull. This lets people build hulls offline and check for hull validity before using them in b2PolygonShape.

This video is a good reference for the 2D quickhull algorithm (my implementation is different): https://www.youtube.com/watch?v=2EKIZrimeuk

Fixes: #671 #728

#711 #732 #731 #729 #723 #718

For #713

lots of confined boxes

This code is migrated from https://github.com/erincatto/box2c

- readme
- build scripts
- maximum linear velocity in world def, fast rotation in body def

Fix migration guide link

Joints were incorrectly using the body bit-set for the overflow color
Sensor shapes now always generate events when created

- hand written SSE2, Neon, and scalar SIMD
- improved performance of SSE2 (x64) and Neon (ARM)
- removed SIMDE library
- SSE2 is the default on x64

Cross platform determinism by implementing custom trigonometry functions
64-bit filter category and mask
Fixed island bug
Added `b2DefaultDebugDraw()`

- renamed smooth segment to chain segment
- addressed several minor issues

#777
#779
#754 
#749 
#782 
#781 
#776

Fixes #774 and more.

Added Intel N100 benchmarks.

Don't using the standard library `rand()`. This way all samples should
be cross platform deterministic.
Simplified clang compile options for determinism.
b2Shape_RayCast now uses b2RayCastInput.

Updated benchmarks for AMD Ryzen when new Windows 11 patch.

#788
#791

Internal arrays using macro generated functions instead of stretchy
buffers.
Added `b2Body_GetWorld`
Added `b2CosSin` and `b2ComputeCosSin`

Wrote a custom capsule versus capsule collider which is more robust than
treating capsules as two face polygons.

add test for null pair task
add manifold to contact begin events

#803 #806

set weld joint reference angle
fixed some angle wrapping bugs

get/set max linear velocity

- optimized ray and shape cast : 2x faster
- update mass options instead of automatic mass
- fixes for 32-bit Windows build
- b2TreeStats for measuring query performance
- reduced tree node size from 48-bytes to 40-bytes
- fixes for 32-bit Neon
- MSVC warning level 4 and fixes
#814, #815, #813, #809

added point overlap test

Added translation and speed functions for prismatic joint.
Fix missing sensor end touch event for fast moving shapes.
Null joint to disable collision between two specific bodies.
World user data.
More consistent functions to make polygons.
Save and load uint64 from ids.

- adjusted AABB margin based on performance testing
- compute rotation between vectors
- added macOS samples to GitHub actions

#833, #835

Sensor and contact end touch events are now generated whenever a contact
is destroyed via user operations, such as:
- destroying a body or shape
- changing filters
- changing body type
- disabling a body
- connecting bodies via joints

Enable hot reloading on MSVC.

It turns out emscripten compiles fine on WSL.
Added benchmark for ragdolls.
Created shared library for code shared by samples, benchmarks, and unit
tests.
Now samples and benchmarks can run off the same code.
Fix solver set leak (thanks Melvyn)
Pinned imgui and glfw versions