Fix restarted workflows not triggering runners

[pavlovic-ivan]

This PR fixes the issue with runners not being created when a workflow is re-run. I've tried to keep changes to ILGPU repo as minimal as possible. Have in mind that there will be a more robust solution developed that i will discuss with @jgiannuzzi. With this PR, support is added when:

• triggering a completely new workflow will trigger runners deployment (runner per cuda job)
• restarting entire workflow when there are failed job will trigger runners deployment (runner per cuda job)
• restarting only failing jobs will trigger runners deployment (runner per cuda job)

However, prerequisite before merging this PR is to create a Github action as described here and here. Have in mind that some of the already defined secrets might change as well. @jgiannuzzi i believe you would be of assistance here for this as i don't have access on that level to ILGPU repo.

@m4rs-mt i did not include any working examples in this PR as we did the demo of the solution several days ago. If needed, i can provide them.

Fixes https://github.com/G-Research/gr-oss/issues/204