Panic when preparing fails

When there are multiple systems and some have not yet started to prepare but one already failed and prepare folds, spread may panic.

```
panic: attempting to unreserve a system that is not reserved: localhost:59315

goroutine 44 [running]:
github.com/snapcore/spread/spread.(*Runner).unreserve(0xc0001cf8c0, 0xc00041c1b0, 0xf)
/home/zyga/projects/spread/spread/runner.go:971 +0x1ad
github.com/snapcore/spread/spread.(*Runner).discardServer(0xc0001cf8c0, 0x8d2080, 0xc000420420)
/home/zyga/projects/spread/spread/runner.go:857 +0x21d
github.com/snapcore/spread/spread.(*Runner).worker(0xc0001cf8c0, 0xc00016c000, 0xc0000da630, 0xc000312b60, 0x1a, 0x1a)
/home/zyga/projects/spread/spread/runner.go:673 +0x12d5
created by github.com/snapcore/spread/spread.(*Runner).loop
/home/zyga/projects/spread/spread/runner.go:232 +0x701
```

Start with spread/runner.go at the reported unreserve, discardServer, and worker paths, then trace how preparation failure is folded when multiple systems are involved. Reproduce the case with systems that have not started preparing alongside a failed system. Done means this scenario no longer panics while cleaning up reservations.

Written by the indexing model from the issue text.