[test] Fix flake in test_dlopen_binary_not_retained (#28041)

In V8, `--concurrent-array-buffer-sweeping` is enabled by default, so a
single synchronous `gc()` call spawns a background `SweepingJob` and
returns before `ArrayBufferSweeper::FinalizeAndDelete` frees dead
`ArrayBuffer` backing stores and updates
`process.memoryUsage().arrayBuffers`. Under CPU contention on CI bots,
the main thread can read `process.memoryUsage().arrayBuffers` before the
background sweep job completes.

Call `gc()` twice before each measurement so that the start of the
second `gc()` (`ArrayBufferSweeper::EnsureFinished()`) waits for the
first `gc()`'s background sweep job to finish.

See: #28002

Author: sbc100

test/test_other.py

@@ -7325,6 +7325,10 @@ def test_dlopen_binary_not_retained(self):
 
       int main() {
         EM_ASM({
+          // Call gc() twice because V8 sweeps ArrayBuffer backing stores
+          // concurrently on a background thread; the start of the second gc()
+          // waits for the first gc()'s ArrayBufferSweeper job to finish.
+          gc();
           gc();
           globalThis.before = process.memoryUsage().arrayBuffers;
         });
@@ -7333,6 +7337,7 @@ def test_dlopen_binary_not_retained(self):
         int (*side_fn)(int) = dlsym(h, "side_fn");
         assert(side_fn && side_fn(1) == 2);
         int retained_mb = EM_ASM_INT({
+          gc();
           gc();
           return (process.memoryUsage().arrayBuffers - globalThis.before) >> 20;
         });