Avoid taking vm barrier in heap_prepare() (ruby#14425)

We can avoid taking this barrier if we're not incremental marking or lazy sweeping.
I found this was taking a significant amount of samples when profiling `Psych.load`
in multiple ractors due to the vm barrier. With this change, we get significant improvements
in ractor benchmarks that allocate lots of objects.

-- Psych.load benchmark --

```
Before:            After:
r:   itr:   time   r:   itr:   time
0    #1:  960ms    0    #1:  943ms
0    ruby#2:  979ms    0    ruby#2:  939ms
0    ruby#3:  968ms    0    ruby#3:  948ms
0    ruby#4:  963ms    0    ruby#4:  946ms
0    ruby#5:  964ms    0    ruby#5:  944ms
1    #1:  947ms    1    #1:  940ms
1    ruby#2:  950ms    1    ruby#2:  947ms
1    ruby#3:  962ms    1    ruby#3:  950ms
1    ruby#4:  947ms    1    ruby#4:  945ms
1    ruby#5:  947ms    1    ruby#5:  943ms
2    #1: 1131ms    2    #1: 1005ms
2    ruby#2: 1153ms    2    ruby#2:  996ms
2    ruby#3: 1155ms    2    ruby#3: 1003ms
2    ruby#4: 1205ms    2    ruby#4: 1012ms
2    ruby#5: 1179ms    2    ruby#5: 1012ms
4    #1: 1555ms    4    #1: 1209ms
4    ruby#2: 1509ms    4    ruby#2: 1244ms
4    ruby#3: 1529ms    4    ruby#3: 1254ms
4    ruby#4: 1512ms    4    ruby#4: 1267ms
4    ruby#5: 1513ms    4    ruby#5: 1245ms
6    #1: 2122ms    6    #1: 1584ms
6    ruby#2: 2080ms    6    ruby#2: 1532ms
6    ruby#3: 2079ms    6    ruby#3: 1476ms
6    ruby#4: 2021ms    6    ruby#4: 1463ms
6    ruby#5: 1999ms    6    ruby#5: 1461ms
8    #1: 2741ms    8    #1: 1630ms
8    ruby#2: 2711ms    8    ruby#2: 1632ms
8    ruby#3: 2688ms    8    ruby#3: 1654ms
8    ruby#4: 2641ms    8    ruby#4: 1684ms
8    ruby#5: 2656ms    8    ruby#5: 1752ms
```

Author: luke-gruber

gc/default/default.c

@@ -985,6 +985,9 @@ total_final_slots_count(rb_objspace_t *objspace)
 #define GC_INCREMENTAL_SWEEP_SLOT_COUNT 2048
 #define GC_INCREMENTAL_SWEEP_POOL_SLOT_COUNT 1024
 #define is_lazy_sweeping(objspace)           (GC_ENABLE_LAZY_SWEEP && has_sweeping_pages(objspace))
+/* In lazy sweeping or the previous incremental marking finished and did not yield a free page. */
+#define needs_continue_sweeping(objspace, heap) \
+    ((heap)->free_pages == NULL && is_lazy_sweeping(objspace))
 
 #if SIZEOF_LONG == SIZEOF_VOIDP
 # define obj_id_to_ref(objid) ((objid) ^ FIXNUM_FLAG) /* unset FIXNUM_FLAG */
@@ -2022,7 +2025,10 @@ static void
 gc_continue(rb_objspace_t *objspace, rb_heap_t *heap)
 {
     unsigned int lock_lev;
-    gc_enter(objspace, gc_enter_event_continue, &lock_lev);
+    bool needs_gc = is_incremental_marking(objspace) || needs_continue_sweeping(objspace, heap);
+    if (!needs_gc) return;
+
+    gc_enter(objspace, gc_enter_event_continue, &lock_lev); // takes vm barrier, try to avoid
 
     /* Continue marking if in incremental marking. */
     if (is_incremental_marking(objspace)) {
@@ -2031,9 +2037,7 @@ gc_continue(rb_objspace_t *objspace, rb_heap_t *heap)
         }
     }
 
-    /* Continue sweeping if in lazy sweeping or the previous incremental
-     * marking finished and did not yield a free page. */
-    if (heap->free_pages == NULL && is_lazy_sweeping(objspace)) {
+    if (needs_continue_sweeping(objspace, heap)) {
         gc_sweep_continue(objspace, heap);
     }