CI: don't have PR runs cancel each other

[axel-h]

Just cancel all older hardware runs for the current PR.

This fixes #65

[axel-h]

See a ci run here, where a new push cancels the older run: https://github.com/seL4/camkes-vm-examples/actions/runs/7705093249?pr=64

[lsf37]

I think that is good. For PRs it makes sense to have the concurrency control at the workflow level and to also cancel the build steps. That'll save runners which are getting scarcer (one PR with a full seL4test HW run on the seL4 repo has 72 jobs currently).

For deployment runs, I kept cancel-in-progress on false to get deployments for those jobs that had already started, but for PRs discarding makes more sense.

I initially thought that the scope of the concurrency control is org-wide, but according to https://github.com/orgs/community/discussions/78332 this is not the case, so using the workflow name is fine.