[DNM] Make self-hosted GPU workflow run again

[ijpulidos]

Solves #126

The only difference I saw was the micromamba setup version, updating it in these changes.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.87%. Comparing base (639b258) to head (5e2602f).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #129   +/-   ##
=======================================
  Coverage   86.86%   86.87%           
=======================================
  Files          11       11           
  Lines        1439     1440    +1     
=======================================
+ Hits         1250     1251    +1     
  Misses        189      189           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ijpulidos]

From the latest actions with self runners they started running again all of a sudden. But we are getting some JSON decoding errors, either way I think the current changes in this branch should be worth it (for the micromamba setup action).

I'll let the self-host test finish in this PR and then change it back to a cron-scheduled automated test.

[mikemhenry]

I am a bit worried the EC2 instance is running but won't ever pick up the job https://github.com/OpenFreeEnergy/feflow/actions/runs/18601132394/job/53046333313?pr=129

[mikemhenry]

Try updating the ami to ami-076a54ed41e67782d which will have the latest cuda drivers

[mikemhenry]

We actually don't have any running so it must have crashed

[ijpulidos]

@mikemhenry Thanks, that's a good tip. Will try running with that ami in the future.

From latest re-run https://github.com/OpenFreeEnergy/feflow/actions/runs/18601132394/job/53091442796?pr=129 I believe the instance is getting stopped abruptly, my guess is that it's hitting some resources limit, I quick test could be to run the tests in serial (not in parallel), and see.