This patch fixes two issues in the HLSL generated to sample an integer cube map.
The first issue was that an inappropriate major axis was selected when sampling
from corners of the cube map. In particular, the added test case demonstrates a
situation where a major axis of length 0 was selected, which lead to an infinite
LOD being calculated. The fix was to adjust inequalities such that always
exactly one of xMajor, yMajor and zMajor is true.
The second issue was that the derivative functions ddx and ddy were being used
on values that depended on the choice of major axis, which is not continuous at
the corners of a cube map. This lead to a finite but incorrect LOD being
calculated. The fix was to make sure major axis choice is not included when
estimating the scaling factor.

Bug: angleproject:3442
Change-Id: Ia3eb8c89f47d1bfdadc18aec989e8cbebc088ec0
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1601515
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>