Concurrent `oasis rofl build` using different deployments

It's common when you make a hotfix that you deploy your ROFL simultaneously to Testnet and Mainnet. When building ORCs concurrently, the second one fails with

Validating compose file...
Preparing stage 2 root filesystem...
Runtime hash: e2e074d03ab2fbacaacb01e6d63ce093fde04e4cc620dd8804e8afbb20390525
Creating squashfs filesystem...
Error: failed to create rootfs image: failed to copy file content: write |1: broken pipe
Pushing ROFL app to OCI repository 'rofl.sh/2ef1ae2d-7f04-495d-be0c-35593853dc55:1787832314'...
Pushing... 100.00% [#################################################] 84.53 MiB

And the old version of .orc is pushed.

Figure out where does the clash happen during the creation of rootfs image and make each deployment use unique tmp directory naming.

ps: Another issue is the race condition between two oasis rofl build instances when updating rofl.yaml. This may not be so trivial. Perhaps it would be more convenient for rofl build command to build ROFL for all deployments out there. Then, we could parallelize it easier.

Start by tracing the oasis rofl build command through rootfs image creation and temporary-directory handling, then reproduce two builds for different deployments concurrently. Done means concurrent builds no longer clash or publish an old .orc; the separate rofl.yaml race is also noted as a possible additional scope.