channel tests: don't reset allocated bytes counter

The ALLOCATED atomic is used by the golang::select2 test
to track the total size of the current heap allocations.
It is reset to 0 during the test, which can cause underflow
if existing allocations are deallocated afterwards.
The golang test this is copied from doesn't reset the alloc stats,
and it doesn't affect the test result, since only the difference
in bytes allocated is asserted.
So don't reset the ALLOCATED counter during the test.

Also only sample ALLOCATED once at the end of the test,
to avoid its value being inconsistent between the 2 loads.

crossbeam-channel/tests/golang.rs

@@ -754,8 +754,6 @@ mod select2 {
         let c = make_unbounded::<i32>();
         let dummy = make_unbounded::<i32>();
 
-        ALLOCATED.store(0, SeqCst);
-
         go!(c, sender(&c, N));
         receiver(&c, &dummy, N);
 
@@ -764,10 +762,9 @@ mod select2 {
         go!(c, sender(&c, N));
         receiver(&c, &dummy, N);
 
-        assert!(
-            !(ALLOCATED.load(SeqCst) > alloc
-                && (ALLOCATED.load(SeqCst) - alloc) > (N as usize + 10000))
-        );
+        let final_alloc = ALLOCATED.load(SeqCst);
+
+        assert!(!(final_alloc > alloc && final_alloc - alloc > N as usize + 10000));
     }
 }