Assertion failure has different influence on GPU and CPU

[FantasyVR]

Describe the bug
Related issue: #6615

When arch=ti.gpu, assertion failure terminates the for loops directly. However, the for loop is not terminated on the CPU backend.

To Reproduce

import taichi as ti 

ti.init(arch=ti.gpu) # when arch=ti.cpu the output is different

@ti.kernel 
def test():
    for i in range(3):
        assert i == 1.0
        print(i)

test()

Log/Screenshots
When arch=ti.gpu the output is shown as:

[Taichi] version 1.3.0, llvm 15.0.4, commit 9c667572, linux, python 3.8.13
[Taichi] Starting on arch=cuda
1

When arch=ti.cpu the output is shown as:

[Taichi] version 1.3.0, llvm 15.0.4, commit 9c667572, linux, python 3.8.13
[Taichi] Starting on arch=x64
0
1
2
3

Additional comments
The final IR results on both backends are the same except that grid_dim and block_dim are different.