Move gc_should_collect_mem_usage() call.

It's probably better to call this inside of gc_collect_main().  That
way, we are not doing the STW from inside _PyObject_GC_Link() function.
This should have no significant performance impact since we hit this
only after the young object count hits the threshold.

Author: nascheme

Python/gc_free_threading.c

@@ -2048,6 +2048,11 @@ gc_should_collect_mem_usage(PyThreadState *tstate)
     GCState *gcstate = &interp->gc;
     int threshold = gcstate->young.threshold;
 
+    if (gcstate->old[0].threshold == 0) {
+        // A few tests rely on immediate scheduling of the GC so we ignore the
+        // extra conditions if generations[1].threshold is set to zero.
+        return true;
+    }
     if (gcstate->deferred_count > threshold * 40) {
         // Too many new container objects since last GC, even though memory
         // use might not have increased much.  This avoids resource
@@ -2096,7 +2101,7 @@ gc_should_collect(PyThreadState *tstate)
         // objects.
         return false;
     }
-    return gc_should_collect_mem_usage(tstate);
+    return true;
 }
 
 static void
@@ -2311,6 +2316,10 @@ gc_collect_main(PyThreadState *tstate, int generation, _PyGC_Reason reason)
         _Py_atomic_store_int(&gcstate->collecting, 0);
         return 0;
     }
+    if (reason == _Py_GC_REASON_HEAP && !gc_should_collect_mem_usage(tstate)) {
+        _Py_atomic_store_int(&gcstate->collecting, 0);
+        return 0;
+    }
     gcstate->frame = tstate->current_frame;
 
     assert(generation >= 0 && generation < NUM_GENERATIONS);