Context

we have a big single step for building the benchmark image and executing the benchmarks on the remote machine, see benchmark-runtime-weights.yml

To increase flexibility, we'd like to:

• add a boolean input parameter to decide if we should build a docker image, or pull it directly from the hub. (Default = build)
• split it into two jobs so that we can restart failed jobs only without starting everything from scratch

✔️ Please set appropriate labels and assignees if applicable.