AArch64: Support SVE operations with encodable immediates

For below case

    for(int i = 0; i < LENGTH; i++) {
      c[i] = a[i] + 2;
    }

AArch64 backend generates SVE add instructions with a immediate move

mov z16.s, #2
add z17.s, z17.s, z16.s

Considering sve has supported basic binary operations with immediate,
we should optimize this pattern, and the new generated vector
instructions should be like

add z16.s, z16.s, #2