Restructure the CI actions to take advantage of idle GPUs #363

Currently we specify which GPUs should a job run on however when do not run parallel jobs on the unused GPUs. We have a single runner so we need to restructure the YAML to do concurrency. For example, if we have test a running on gpus 1,2 and test b running on gpus 3, 4, 5, 6, 7, 8 then on our 8 gpu system these two should run in parallel but atm they run serially. Restructure the CI actions to take advantage of idle GPUs