Use add instead of or

Subject: https://github.com/rfalke/decompiler-subjects/tree/master/from_holdec/dmi/transformation/ia32_elf/subject.exe

Source:
```
// Address range: 0x8048810 - 0x8048904
int32_t advanced_1_unrolled_loop(void) {
...
for (int32_t i = 48; i < 57; i += 2) {
// 0x8048859
*v2 = i;
putchar((int32_t)&g1);
*v2 = i | 1;
putchar((int32_t)&g1);
}
...
```

Objdump:
```
8048859: 83 ec 0c sub $0xc,%esp
804885c: 53 push %ebx
804885d: e8 be fa ff ff call 8048320
8048862: 8d 43 01 lea 0x1(%ebx),%eax
8048865: 83 c3 02 add $0x2,%ebx
8048868: 89 04 24 mov %eax,(%esp)
804886b: e8 b0 fa ff ff call 8048320
8048870: 83 c4 10 add $0x10,%esp
8048873: 83 fb 3a cmp $0x3a,%ebx
8048876: 75 e1 jne 8048859
```

So it is a `lea` in the binary and not an `or`. retdec converts this to this `|1` (which is correct because we start even) but makes the output harder to understand.

Start with the linked decompiler subject and compare the source around advanced_1_unrolled_loop with the objdump at 0x8048859–0x8048876. Trace where the lea instruction becomes `| 1`, then identify the relevant transformation or output tests; done means the generated expression reflects the add/lea operation while preserving correctness.

Written by the indexing model from the issue text.