Running with --no-graphics and Visual Observations fails silently

[jefffhaynes]

Running mlagents-learn with the --no-graphics options silently causes visual observation training to fail. This issue is documented here. However, unless you happen to read this documentation, it is possible to spend weeks running simulations that will simply never converge with no feedback that there is no meaningful training happening.

• Unity Version: Unity 6000.1.0f1
• OS + version: Windows 11
• ML-Agents version: 1.2.0.dev0