Regression in shader discard CTS test failures

[billhollings]

There has been a recent regression in CTS fragment_shader_interlock discard tests. An additional 32 tests are failing.

The newly failing tests all seem to be those of type1xaa (one sample per pixel).

There are an existing set of 45 4xaa tests that were already failing, so it might be best to bundle all 77 of them altogether in a fix, rather than just addressing the 1xaa tests.

The full set of 576 fragment_shader_interlock tests is in this file: fragment-shader-interlock.txt

The full set of 77 failing fragment_shader_interlock tests is in this file: fragment-shader-interlock.txt-fails.txt

And the set of 32 newly-failing 1xaa tests that make up the regression are in this file: fragment-shader-interlock.txt-new-regression-fails.txt

@aitor-lunarg I'm wondering if this is related to your recent PR #2248, covering discard side-effects.

[aitor-lunarg]

I just checked and it does make it fail. I'll provide a PR to revert that part of the change and investigate later