Set torch device from commandline

[chriselion]

Proposed change(s)

The torch device string can now be set from the commandline. The default behavior is the same as before: it will use cuda if it's detected, or cpu otherwise. This allows overriding the default logic (e.g. force cpu even if you support cuda), or specifying a device index if e.g. you have two different GPU types.

Also logs the device type so that it's obvious which is being used.

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

https://forum.unity.com/threads/ml-agent-server-build-headless-with-visual-observations-and-multiple-gpus.1042456/
#4876

Types of change(s)

• New feature

Checklist

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

[dongruoping]

It would be good to also mention this option somewhere in to doc

[ervteng]