Fix "zero-sized fn" nightly warnings

[gkoz]

Example:

src/signal.rs:93:70: 93:79 warning: `extern "C" fn(&core::cell::RefCell<Box<core::ops::FnMut() -> glib::source::Continue + 'static>>) -> i32 {signal::trampoline}` is now zero-sized and has to be cast to a pointer before transmuting to `core::option::Option<unsafe extern "C" fn(*mut libc::c_void) -> i32>`, #[warn(transmute_from_fn_item_types)] on by default

src/signal.rs:93         glib_ffi::g_idle_add_full(glib_ffi::G_PRIORITY_DEFAULT_IDLE, transmute(trampoline),

                                                                                      ^~~~~~~~~

src/signal.rs:93:70: 93:79 warning: this was previously accepted by the compiler but is being phased out; it will become a hard error in a future release!

src/signal.rs:93:70: 93:79 note: for more information, see issue #19925 <https://github.com/rust-lang/rust/issues/19925>

We should make our extern fns match the expected signatures to make transmutes unnecessary.

[GuillaumeGomez]

Can't we just cast as *mut c_void? It would do the same after all.

[gkoz]

Transmuting fn is too fragile anyway. More error-prone than transmuting or casting individual arguments.

[gkoz]

Um, well, speaking of signals there may not be actual signatures to adhere to. I'm not sure now. So this issue needs some investigation.

[EPashkin]

rust-lang/rust#19925 stated that we can do even transmute(size_changed_trampoline as usize).
As for @gkoz concern about transmuting individual arguments IMHO it not applied here, except for &Box<Fn> argument that really not error-prone.
May be better solution will be pass some structure with some id and function pointer that will be checked before using function but it needed some storage with unclear rule for erasing elements.
As we need transmute trampoline anyway IMHO better add cast sooner.

[gkoz]

It looks like we'll have to use the as usize workaround for signal trampolines because there's no other choice there. Callbacks with defined types (like GtkTreeModelFilterVisibleFunc) and g_idle_add can avoid transmuting fns.

[EPashkin]

I add PR for gtk and glib then.

[EPashkin]

Also calling g_idle_add_full and others in glib also needed casting.

[gkoz]

Looks like all warnings are gone. Thanks @EPashkin