Physics plugin execution scheme documentation does not match code

[brettle]

The execution scheme document says that dSpaceCollide is called before webots_physics_step():

webots/docs/reference/execution-scheme.md

Lines 12 to 13 in edf2c18

	S3 --> S4["dSpaceCollide(space)</br>is invoked which eventually causes multiple callbacks to</br>webots_physics_collide()"]
	S4 --> S5["webots_physics_step()</br>is called to allow the user to setup additional forces, etc."]

However in the code, it looks like webots_physics_step() is called before collisions. See:

webots/src/webots/engine/WbSimulationWorld.cpp

Lines 241 to 249 in edf2c18

	if (mPhysicsPlugin)
	mPhysicsPlugin->step();
	WbRadio::runPlugin(timeStep);
	if (log) {
	log->stopMeasure(WbPerformanceLog::PRE_PHYSICS_STEP);
	log->startMeasure(WbPerformanceLog::PHYSICS_STEP);
	}
	mCluster->step();

webots_physics_step() is called as a result of line 242 and collisions are done during the execution of line 249 (in particular, mCluster->step() calls dWorldStepAndSpaceCollide().

I've marked this as a documentation bug for now, but it's not clear if the current order in the code is intentional/desirable. In particular, the current order precludes a physics plugin from influencing the physics after all of the contact joints have been created.

[omichel]

I agree there is a mismatch here and we should probably fix the documentation in a first step.

Then, it should be investigated, probably in a PR, whether it is possible and desirable to change this order.
The PR should demonstrate that changing this order doesn't break anything that can't be repaired and should argue the benefits of such a change.