[Bug]: nonut (unsigned) comparison, division, +, -, * are computed as signed int

[leo-aa88]

Description

nonut (unsigned) integer values are stored correctly (%u prints the right bits), but every binary-operation path in handle_binary_operation except OP_MOD evaluates the operands as signed int. Comparisons, division, +, -, and * therefore give wrong answers for any nonut rizz value whose high bit is set (i.e. > INT_MAX). The is_unsigned modifier never reaches these nodes, so the value is treated as a negative signed int.

This is distinct from the already-fixed #25 (which was specifically nonut modulo).

Minimal reproduction code

skibidi main {
    nonut rizz a = 3000000000;
    edgy (a > 5) {
        yapping("big");
    } amogus {
        yapping("small");
    }
    bussin 0;
}

Also, division:

skibidi main {
    nonut rizz a = 3000000000;
    nonut rizz b = 2;
    yapping("%u", a / b);   // expect 1500000000
    bussin 0;
}

Expected behavior

• Comparison: 3000000000 > 5 is true → prints big.
• Division: 3000000000 / 2 → 1500000000.

Actual behavior

$ ./brainrot cmp.brainrot
small
$ ./brainrot div.brainrot
3647483648

3000000000 reinterpreted as signed int is -1294967296 (negative, so < 5); -1294967296 / 2 = -647483648, which as unsigned prints 3647483648.

Brainrot version / commit

19b9563

Operating system

Ubuntu on WSL2 (Linux 5.15)

Additional context

Root cause: ast.c, handle_binary_operation. promoted_type is only ever VAR_SHORT/VAR_INT/VAR_FLOAT/VAR_DOUBLE, and the VAR_INT arms for OP_LT/GT/LE/GE/EQ/NE, OP_DIVIDE, OP_PLUS/MINUS/TIMES operate on signed int. Only OP_MOD consults node->modifiers.is_unsigned — and the in-code comment in the OP_DIVIDE arm already notes that is_unsigned "is never set on these nodes as things stand," so even that handling is effectively dead. The fix needs to first propagate is_unsigned onto binary-op nodes, then add unsigned branches (using unsigned int arithmetic) to each arm.