In workflows like https://github.com/llvm/llvm-project/actions/runs/17234316420/job/48895489095, sccache is failing build actions.

It seems like the server is dying and then the clients are not failing gracefully and falling back to local compilation. I can reproduce this locally by killing the server. Not sure why the server is dying. I think there are two things we need to do here:

1. Figure out why the server is dying. I'm hoping we can get some info by setting SCCACHE_LOG=info and dumping the log file into the artifacts.
2. Make sccache fail gracefully when it cannot connect to the server.