builtins: overflow arithmetic perf optimization

[matu3ba]

builtins: overflow arithmetic perf optimization

change signature of arithmetic operations @addwithOverflow,
@subWithOverflow, @mulWithOverflow, shlWithOverflow from
@operation(comptime T: type, a: T, b: T, result: *T) bool
to
@operation(comptime T: type, a: T, b: T) anytype
with anytype being a tuple
struct { res: T, ov: bool }

This removes the pointer store and load for efficiency of codegen.
Comptime operation is accordingly kept in sync.

closes #10248

[matu3ba]

I am abit stuck at figuring out how to construct a composite type in Air for https://github.com/ziglang/zig/pull/10854/files#diff-23f8450a3403f774f20a81390ea9072846dc675ee17a679f715b26b22c86b591R8132, since the result should be a struct {res: T, ov:bool} as shown here https://github.com/matu3ba/zorro/blob/44a3ac0aa37bacb54623708ea3437462c2d20492/src/crt/addo_noptr.zig#L17.
- const ptr = sema.resolveInst(extra.ptr); needs to be removed, but I need to figure out how to generate the struct in Air.
It uses additional data besides what is in Zir and Ast.
For the builtins I found no "stuff that allocates further data", so a hint where to look would be nice.
Otherwise I will debug stage1 (and Sema) for aforementioned example, which should reveal how to do it.