stage2: add error for returning pointer to local variable

[Vexu]

The check does not apply to inline functions and wont apply to functions called at comptime depending on #11152.
Most simple cases are caught but it is easy to bypass if needed.

Some cases might be too aggressive but they are easy to remove if deemed so.

Closes #2646

[andrewrk]

One thing we need to do here is make sure we can manage to make the language specification compatible with this compile error.

cc @SpexGuy

Also related: #3180
I have an idea to try to solve use-after-free of locals at runtime by doing escape analysis, secretly allocating escaped things on the heap with GPA, then freeing them on return. I have not yet done a proof-of-concept for this idea.

Consider this problem:

fn foo() *i32 {
    return undefined;
}

This is valid zig code. It's effectively the same thing as returning the address of a local. We'll need to get creative with the lang spec to make this a compile error.

[Jarred-Sumner]

This is valid zig code

I do this currently when implementing a large interface for something function by function

Just so the code compiles even though the implementation is incomplete

[andrewrk]

@Jarred-Sumner why not instead @panic("TODO")? Or even just unreachable?

[andrewrk]

Closing since we currently have no workable plan to make this a legal compile error according to the lang spec. We can revisit this patch when dealing with #2646 which now has a link to this closed PR. The next step towards landing this patch is coming up with a lang spec proposal that allows this compile error. Please make such proposals as comments on #2646.