[MLA-1129] Clear ActionBuffers before Heuristic calls

[chriselion]

Proposed change(s)

Reset the ActionBuffers before calls to Agent.Heuristicand IHeuristicProvider.Heuristic. Whether or not the actions carried over between steps before was unspecified, not they're always initialized to zero.

Also clean up cases in the examples where we needed to work around the old behavior.

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

https://jira.unity3d.com/browse/MLA-1129
https://jira.unity3d.com/browse/MLA-1123

Types of change(s)

• Code refactor
• Breaking change

Checklist

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

[chriselion]

@surfnerd Anything that needs to change in the input system actuators? I had a quick look and it didn't look like it.

[surfnerd]

@surfnerd Anything that needs to change in the input system actuators? I had a quick look and it didn't look like it.

I don't think so. The input actuator just fills the ActionBuffers with whatever input state exists in the Heuristic method every frame. So there was no need to do any sort of clear in that case.