improve apa102 display fcn

Actually for APA102 LEDs, the start-frame needs to be zeros of 32bit size.
For the end-frame n/2 bits needs to be written (n is size of apa102 strip.
This improvement reduces the bytes needed to be written resulting in notable increase of speed.

luma/led_matrix/device.py

@@ -420,13 +420,13 @@ def display(self, image):
         assert image.size == self.size
         self._last_image = image.copy()
 
-        # Send zeros to reset, then pixel values then zeros at end
+        # Send 32 zero-bits to reset, then pixel values then n/2 zero-bits at end
         sz = image.width * image.height * 4
-        buf = bytearray(sz * 3)
+        buf = bytearray(4 + sz + round(image.width * image.height/8/2))
 
         m = self._mapping
         for idx, (r, g, b, a) in enumerate(image.getdata()):
-            offset = sz + m[idx] * 4
+            offset = 4 + m[idx] * 4
             brightness = (a >> 4) if a != 0xFF else self._brightness
             buf[offset] = (0xE0 | brightness)
             buf[offset + 1] = b