Chaining uadd.with.overflow should optimize to or disjoint

[scottmcm]

The following code is useful for doing bigint addition, where you need to add the the digits from both numbers but also add a carry https://cpp.godbolt.org/z/c7PcWqbz7

bool uadd_with_carry(unsigned a, unsigned b, bool c, unsigned* out) {
    unsigned sum1;
    bool c1 = __builtin_uadd_overflow(a, b, &sum1);
    unsigned sum2;
    bool c2 = __builtin_uadd_overflow(sum1, (unsigned)c, &sum2);
    
    *out = sum2;
    return c1 | c2;
}

That c1 | c2 stays an or i1, but it would be legal for it to be or disjoint i1: https://alive2.llvm.org/ce/z/BMwvqN

(This is a C++ translation of Rust's https://doc.rust-lang.org/std/primitive.u32.html#method.carrying_add, cc rust-lang/rust#85532)

[AmrDeveloper]

@dtcxzyw If it's confirmed and has priority I can start working on it

[dtcxzyw]

Similar with #118164, I am just wondering if we can get better codegen with or disjoint i1.

[scottmcm]

I ended up just updating rust's implementation of this method to disjoint manually, so you can close this if you want.