SF: fix off-by one in luma mean calculation

There was an off by one error in luma mean calculation that could
lead to segfault. Flaw was figured out in writing the 5 attached
unit tests for the function. Condition would be obscure in the wild,
but an exactly half-white, half-another-color in the sampled region
could present condition.

Fixes: 129858549
Test: libsurfaceflinger_unittest (5 new tests)

Change-Id: I920cd9cac15122178ec9258e33c9bc35b1bb9357