[Bug] ContinueStmt failed to trigger on CUDA backend

[jim19930609]

Issue from forum: https://forum.taichi-lang.cn/t/topic/4342

Reproduce

import taichi as ti
ti.init(arch=ti.gpu)
img = ti.field( ti.i32, (2,2))
@ti.kernel
def K():
    for i,j in img:
        img[i,j] = 0
        if  i> 0  or j> 0:continue
        img[i,j] = 1

    for i,j in img:
        print(i,j,img[i,j])

img.fill(2)
K()

Output

[CUDA - incorrect]

[Taichi] version 1.7.0, llvm 15.0.4, commit 5712f644, linux, python 3.10.6
[Taichi] Starting on arch=cuda
0 0 1
0 1 2
1 0 2
1 1 2

[CPU - reference]

[Taichi] version 1.7.0, llvm 15.0.4, commit 5712f644, linux, python 3.10.6
[Taichi] Starting on arch=x64
0 0 1
0 1 0
1 0 0
1 1 0