Make SPI bit-banging take less time.

The spi_bus.py loop puts 0 and 1 bits on the SPI bus by flipping, separated by time.sleep(1e-6). Which of course take ~75us.

This particular loop should probably be in cython/c, but for now use a null method, which takes about 1 us on a Pi3.
Also, this calls GPIO.output(), which obviously takes >1us itself. Will try removing all sleeps, but that will need more testing.