Develop modify stepping logic

[vincentpierre]

Proposed change(s)

Making changes to the step counting in both the Academy and the Agent.
The Academy will first set status, then tick its step count, then tick all agents step counts.
Collect observations and Act happen as usual.
At the end of the loop, the Agent resets if max step was reached.

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

Is continuation of #3432
Jira is MLA-619

Types of change(s)

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

Checklist

• I have added tests that prove my fix is effective or that my feature works
• I have added updated the changelog (if applicable)
• I have added necessary documentation (if applicable)
• I have updated the migration guide (if applicable)

Other comments

[surfnerd]

what does the step count look like for an episode?

It looks like if maxStep = 5 the step count will look like:

AgentStep = 1
  Agent Observe; Agent Act;
AgentStep = 2
  Agent Observe; Agent Act;
AgentStep = 3
  Agent Observe; Agent Act;
AgentStep = 4
  Agent Observe; Agent Act;
AgentStep = 5
  Agent Observe; Agent Act; Agent Reset;

Is my understanding correct?

[vincentpierre]

Is my understanding correct?

Yes, this is what I am aiming at

[surfnerd]

seems to make sense. Should there be a test for this?

[chriselion]

Looks good, just add something in the changelog.