test/pylib: scylla_cluster: don't leak server if stopping it fails

`ScyllaCluster.server_stop` had this piece of code:
```
        server = self.running.pop(server_id)
        if gracefully:
            await server.stop_gracefully()
        else:
            await server.stop()
        self.stopped[server_id] = server
```

We observed `stop_gracefully()` failing due to a server hanging during
shutdown. We then ended up in a state where neither `self.running` nor
`self.stopped` had this server. Later, when releasing the cluster and
its IPs, we would release that server's IP - but the server might have
still been running (all servers in `self.running` are killed before
releasing IPs, but this one wasn't in `self.running`).

Fix this by popping the server from `self.running` only after
`stop_gracefully`/`stop` finishes.

Make an analogous fix in `server_start`: put `server` into
`self.running` *before* we actually start it. If the start fails, the
server will be considered "running" even though it isn't necessarily,
but that is OK - if it isn't running, then trying to stop it later will
simply do nothing; if it is actually running, we will kill it (which we
should do) when clearing after the cluster; and we don't leak it.

Closes #12613

test/pylib/scylla_cluster.py

@@ -757,11 +757,14 @@ class ScyllaCluster:
         if server_id not in self.running:
             return ScyllaCluster.ActionReturn(success=False, msg=f"Server {server_id} unknown")
         self.is_dirty = True
-        server = self.running.pop(server_id)
+        server = self.running[server_id]
+        # Remove the server from `running` only after we successfully stop it.
+        # Stopping may fail and if we removed it from `running` now it might leak.
         if gracefully:
             await server.stop_gracefully()
         else:
             await server.stop()
+        self.running.pop(server_id)
         self.stopped[server_id] = server
         return ScyllaCluster.ActionReturn(success=True, msg=f"{server} stopped")
 
@@ -781,8 +784,10 @@ class ScyllaCluster:
         self.is_dirty = True
         server = self.stopped.pop(server_id)
         server.seeds = self._seeds()
-        await server.start(self.api)
+        # Put the server in `running` before starting it.
+        # Starting may fail and if we didn't add it now it might leak.
         self.running[server_id] = server
+        await server.start(self.api)
         return ScyllaCluster.ActionReturn(success=True, msg=f"{server} started")
 
     async def server_restart(self, server_id: ServerNum) -> ActionReturn: