Warn when control flow reaches end of a `_Noreturn` function

[michael-schwarz]

Since goblint/cil#58, CIL supports the _Noreturn attribute.

As remarked by @sim642 on Slack, we should also do something with _Noreturn in Goblint and emit a warning when control flow does indeed reach the end of such a function (i.e. return node is not dead).

This probably also requires investigating which standard library functions terminate execution (e.g. exit(...) and friends).

[sim642]

I think we already handle the usual suspects as non-returning:

analyzer/src/analyses/base.ml

Lines 2188 to 2205 in e81b1a1

	| `Unknown "exit" -> raise Deadcode
	| `Unknown "abort" -> raise Deadcode
	| `Unknown "__builtin_unreachable" when get_bool "sem.builtin_unreachable.dead_code" -> raise Deadcode (* https://github.com/sosy-lab/sv-benchmarks/issues/1296 *)
	| `Unknown "pthread_exit" ->
	begin match args with
	| [exp] ->
	begin match ThreadId.get_current (Analyses.ask_of_ctx ctx) with
	| `Lifted tid ->
	let rv = eval_rv (Analyses.ask_of_ctx ctx) ctx.global ctx.local exp in
	ctx.sideg (V.thread tid) (G.create_thread rv);
	(* TODO: emit thread return event so other analyses are aware? *)
	(* TODO: publish still needed? *)
	publish_all ctx `Return (* like normal return *)
	| _ -> ()
	end;
	raise Deadcode
	| _ -> failwith "Unknown pthread_exit."
	end

Of course I guess a modern standard library header would also have them with _Noreturn (or equivalent). Actually it might be even worth having a sem option, whether to trust _Noreturn attributes (and make them all raise Deadcode just based on that). While doing that we could still warn about control flow reaching the end of the function within its implementation.
This might avoid a cascade of _Noreturn warnings if we don't handle the innermost root cause.