Errors when scaling up cluster no longer propagate to client side

**Describe the issue**:

Since #8233, scaling up a cluster with a worker plugin that produces an error no longer propagates that error onto the client side. _Creating such a cluster_ with a non-zero number of workers initially does produce an error.

Is this intended behaviour? And if so, is there a new blessed way of propagating errors when scaling up a cluster?

**Minimal Complete Verifiable Example**:

Running with distributed `2023.10.0-21-gb62b7005`

```python
from distributed import LocalCluster

class ErroringPlugin:
def setup(self, worker=None):
raise RuntimeError()

if __name__ == "__main__":
# This does not error
with LocalCluster(n_workers=0, plugins={ErroringPlugin()}) as cluster:
cluster.scale(1)
cluster.sync(cluster._correct_state)

# This does
with LocalCluster(n_workers=1, plugins={ErroringPlugin()}) as cluster:
pass
```