[AMDGPU] Fix typo in cttz_zero_undef(x) -> cttz(x) fold test

v_cttz_zero_undef_i64_with_select should be selecting '64' for the x != 0 case instead of '32' like we just did in the previous 'v_cttz_zero_undef_i32_with_select' test.

Noticed by accident because it was causing some weird regressions....

Differential Revision: https://reviews.llvm.org/D125612