`CriticalSection` should not be `Send` or `Sync`

[matanui159]

Allowing the CriticalSection to be both Send and Sync allows multiple threads to gain access to a resource by sharing it between threads. For example with std this could be done without any unsafe code:

use std::{cell::Cell, thread};
use critical_section::{CriticalSection, Mutex};

static LIST: Mutex<Cell<Vec<i32>>> = Mutex::new(Cell::new(Vec::new()));

fn push_to_list(cs: CriticalSection, iter: impl IntoIterator<Item = i32>) {
    for item in iter {
        let mut list = LIST.borrow(cs).take();
        list.push(item);
        LIST.borrow(cs).set(list);
        thread::yield_now();
    }
}

fn main() {
    critical_section::with(|cs| thread::scope(|scope| {
        scope.spawn(|| {
            push_to_list(cs, 0..100);
        });
        push_to_list(cs, 100..200);
    }));

    critical_section::with(|cs| {
        println!("{:?}", LIST.borrow(cs).take());
    });
}

This causes undefined behaviour and often crashes:

csbug(12007,0x16b3cb000) malloc: *** error for object 0x11f606290: pointer being realloc'd was not allocated
csbug(12007,0x16b3cb000) malloc: *** set a breakpoint in malloc_error_break to debug
fish: Job 1, 'cargo run' terminated by signal SIGABRT (Abort)

[jamesmunns]

This was discussed briefly in chat, it seems like we should ensure that the CS tokens are !Send and !Sync.

This could likely be done in a 1.2 release (and yank older versions), as it is a fix for soundness, and any uses like this would be unsound.