Fix hanging direct scheduler+ssh

The fix is very simple: in the ssh transport, to emulate 'chdir',
we keep the current directory in memory, and we prepend every command
with a `cd FOLDER_NAME && ACTUALCOMMAND`.

One could put `;` instead of `&&`, but then if the folder does not
exist the ACTUALCOMMAND would still be run in the wrong folder, which is
very bad (imagine you are removing files...).

Now, in general this is not a problem. However, the direct scheduler
inserts a complex-syntax bash command to run the command in the background
and immediately get the PID of that process without waiting.
When combined with SSH, this hangs until the whole process is completed, unless
the actual command is wrapped in brackets.

A simple way to check this is running these two commands, that reproduce
the issue with plain ssh, without paramiko:

This hangs for 5 seconds:
```
ssh localhost 'cd tmp && sleep 5 > /dev/null 2>&1 & echo $!'
```

This returns immediately, as we want:
```
ssh localhost 'cd tmp && ( sleep 5 > /dev/null 2>&1 & echo $! )'
```

aiida/transports/plugins/ssh.py

@@ -1265,7 +1265,7 @@ def _exec_command_internal(self, command, combine_stderr=False, bufsize=-1):  #
 
         if self.getcwd() is not None:
             escaped_folder = escape_for_bash(self.getcwd())
-            command_to_execute = (f'cd {escaped_folder} && {command}')
+            command_to_execute = (f'cd {escaped_folder} && ( {command} )')
         else:
             command_to_execute = command