Fix: join the control thread before closing the CAN socket

[tendant]

DMChainCanInterface.close() sets running = False and closes the CAN socket in the same breath, without waiting for the control loop to notice:

def close(self) -> None:
    self.running = False
    self.motor_interface.close()

A thread already inside _set_commands then sends on a socket that has just been closed, and dies on every clean shutdown:

Robot closed with all torques set to zero.
ERROR:root:1th motor at DMChainCanInterface(channel=can0) failed with info [2, 'DM4340']
DM Error in control loop: file descriptor cannot be a negative integer (-1)
Exception in thread Thread-1 (_set_torques_and_update_state):
  File "i2rt/motor_drivers/can_interface.py", line 54, in _send_message_get_response
    self.bus.send(message)
ValueError: file descriptor cannot be a negative integer (-1)

It could not be joined, because start_thread() assigned the handle to a local and dropped it. Keep it, and join it in close() with a timeout so close() stays bounded if the loop is wedged — in that case behaviour is unchanged.

MotorChainRobot.close() already does this for its own thread one level up, so this brings the motor chain in line with the pattern beside it.

Cosmetic: it fires after torques are zeroed, so the arm is safe either way. But it is noisy enough to bury anything else printed around a shutdown, and it reproduces on every disconnect — including a plain connect/read/disconnect with no motion commanded.