fix(antenna): Retract and Calibrate report the elements moving

An operator reported that retracting a SteppIR shows nothing, while an
Ultrabeam visibly does. The cause is not the display: both antennas feed
one `moving` flag and the widget renders it the same way.

moveCmdAt — the bridge that reports motion from the moment a move is
COMMANDED, before the controller's own poll can confirm it — was set
only by SetFrequency. Retract and Calibrate write their frame directly
and armed nothing, so three things were missing at once: nothing said
the elements were moving, the poll stayed on its two-second idle cadence
instead of speeding up to 250 ms to watch, and the transmit inhibit was
not engaged while the elements travelled. That last one is the reason
this is not cosmetic.

The Ultrabeam's Retract armed nothing either. It was less visible there
because that controller reports element lengths, which count down on
screen; a SteppIR reports none at all, so the retract looked inert. The
inhibit was equally absent.

The window becomes a deadline rather than a timestamp, so the caller can
say how long a bridge it needs: three seconds for a tune, ten for a
retract or a calibrate. Those are tens of seconds of travel, and a
retract drops the controller out of AUTOTRACK where its motor-bit
reporting is less predictable. Still bounded — an antenna that never
reports motion must not latch the inhibit on for ever — and markMoving
only ever extends, so a tune issued just after a retract cannot cut the
retract's bridge down to its own.

Four tests, including that a command which failed to reach the
controller does not claim the antenna is moving.

internal/ultrabeam/ultrabeam.go

@@ -894,9 +894,20 @@ func (c *Client) SetDirection(direction int) error {
 }
 
 // Retract retracts all elements (command 2)
+//
+// Reports motion at once, like a frequency change does: this is the longest
+// move the antenna makes, and the flag it sets is what inhibits the
+// transmitter while the elements travel. The element lengths counting down
+// made the omission less visible here than on a SteppIR, which reports no
+// lengths at all — but the inhibit was equally missing.
 func (c *Client) Retract() error {
-	_, err := c.sendCommand(CMD_RETRACT, nil)
-	return err
+	if _, err := c.sendCommand(CMD_RETRACT, nil); err != nil {
+		return err
+	}
+	c.statusMu.Lock()
+	c.moveCmdAt = time.Now()
+	c.statusMu.Unlock()
+	return nil
 }
 
 // ModifyElement modifies element length (command 12)