[WIP] [Fix] Fixing collect observation called on done

[vincentpierre]

Proposed change(s)

@ervteng noticed that "If you put a GoalSensor/VectorSensorComponent on a dodgeball agent, and add to it in Collectobservations, it will throw a warning when the agent dies"
Investigation showed that the code

            if (collectObservationsSensor != null)
            {
                // Make sure the latest observations are being passed to training.
                collectObservationsSensor.Reset();
                using (m_CollectObservationsChecker.Start())
                {
                    CollectObservations(collectObservationsSensor);
                }
            }

is problematic when using a VectorSensor because only the collectObservationsSensor will be reset BUT both the VectorSensor and collectObservationsSensor will have new observations (because CollectObservations is called).
The proposed solution is to replace the call to collectObservationsSensor.Reset(); with UpdateSensors which is what we do in SendInfoToBrain() already. We also remove the check if (collectObservationsSensor != null) because there could be a Vector Sensor without a collectObservationsSensor.

I also modified the test of the stacking sensor to fit this new logic.

Useful links (Github issues, JIRA tickets, ML-Agents forum threads etc.)

Slack message

Types of change(s)

• Bug fix
• New feature
• Code refactor
• Breaking change
• Documentation update
• Other (please describe)

Checklist

• Added tests that prove my fix is effective or that my feature works
• Updated the changelog (if applicable)
• Updated the documentation (if applicable)
• Updated the migration guide (if applicable)

Other comments

[dongruoping]

We should add tests to make sure the last observation is not missing nor doubled.