Missed Optimization: fold icmp samesign ugt (x - 18), 7 (with nsw subtract) into icmp sgt x, 25

```
define i1 @src(i32 %arg0) {
  %v0 = add nsw i32 %arg0, -18
  %v1 = icmp samesign ugt i32 %v0, 7
  ret i1 %v1
}

define i1 @tgt(i32 %arg0) {
  %v1 = icmp sgt i32 %arg0, 25
  ret i1 %v1
}
```

alive2: https://alive2.llvm.org/ce/z/HsxeWy
godbolt: https://godbolt.org/z/Ts68hfhKn
Pattern found in: https://github.com/dtcxzyw/llvm-opt-benchmark/blob/main/bench/ffmpeg/optimized/mlpenc.ll