kraid: Lower OpCopy to IADD whenever possible

OpIAdd runs on the CVT pipe and OpShiftLop runs on the SFU pipe.  The
SFU pipe is generally slower so we want to use IADD when we can.

Shader-db results on v13:

    total instrs in shared programs: 641536 -> 641536 (0.00%)
    instrs in affected programs: 0 -> 0
    helped: 0
    HURT: 0

    total cvt in shared programs: 1473.20 -> 1474.08 (0.06%)
    cvt in affected programs: 27.80 -> 28.69 (3.18%)
    helped: 0
    HURT: 109
    HURT stats (abs)   min: 0.007811999999999986 max: 0.023438000000000014 ◆: 0.01 ▒: 0
    HURT stats (rel)   min: 1.39% max: 33.33% ◆: 5.81% ▒: 5.56%
    95% mean confidence interval for cvt value: 0.01 0.01
    95% mean confidence interval for cvt %-change: 4.66% 6.96%
    Cvt are HURT.

    total sfu in shared programs: 2947.62 -> 2940.56 (-0.24%)
    sfu in affected programs: 33.50 -> 26.44 (-21.08%)
    helped: 109
    HURT: 0
    helped stats (abs) min: 0.0625 max: 0.1875 ◆: 0.06 ▒: 0
    helped stats (rel) min: 3.12% max: 100.00% ◆: 43.20% ▒: 33.33%
    95% mean confidence interval for sfu value: -0.07 -0.06
    95% mean confidence interval for sfu %-change: -49.24% -37.16%
    Sfu are helped.

Pipeline-db results on v13:

    Totals:
    Cycle count: 155729.08586799976 -> 155728.96086799976 (-0.00%)
    Estimated normalized CVT cycles: 4297.492723000092 -> 4305.6333350000905 (+0.19%)
    Estimated normalized SFU cycles: 8565.8125 -> 8500.6875 (-0.76%)

    Totals from 66 (0.89% of 7436) affected shaders:
    Cycle count: 24690.875 -> 24690.75 (-0.00%)
    Estimated normalized CVT cycles: 765.8281290000002 -> 773.9687409999997 (+1.06%)
    Estimated normalized SFU cycles: 1208.4375 -> 1143.3125 (-5.39%)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/44722>