glsl-1.30: Test uint to float conversion of sign extended value
Intel compiler has a very old optimization for various combination
of u2f or i2f of an extract_* source. However, the variations that
do u2f of an extract_i* are incorrect.
Intel drivers with the known bug will produce blue.
v2: Use 0x000000ff in the int8 test to match the int16 test. Suggested
by Caio.
v3: Update header comment to explain why two possible incorrect values
are checked. Suggested by Caio.
Part-of: <https://gitlab.freedesktop.org/mesa/piglit/-/merge_requests/883>