altos: Correct incremental rotation computation

altos: Correct incremental rotation computation

Trying to compute the combined rotation by taking the x/y/z rotations
as a vector is a good approximation, but not accurate enough for our
application given the large angles we sometimes see.

Instead, use a correct-but-expensive function with a pile of
transcendental function calls. The STM32L seems to be fast enough at least...

Signed-off-by: Keith Packard <keithp@keithp.com>