ELUD: Do not quarantine direct-mapped allocations

Direct-mapped allocations are generally very large and zapping
them takes long time and causes browser janks. Plus, they're
unmapped when free'ed, so there should be less (or even negative)
value to quarantine direct-mapped allocations.

c.f. https://crrev.com/c/5952195

Bug: 40944045
Change-Id: I171c53f6476e3840ca6af932becff5bb8b8d42a4
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5963883
Commit-Queue: Yuki Shiino <yukishiino@chromium.org>
Reviewed-by: Takashi Sakamoto <tasak@google.com>
Reviewed-by: Sergei Glazunov <glazunov@google.com>
Cr-Commit-Position: refs/heads/main@{#1374540}


CrOS-Libchrome-Original-Commit: 0edb68d842aa463df2152ac4155bbad17a2ca59e

Author: yuki3

base/allocator/partition_allocator/src/partition_alloc/partition_root.h

@@ -752,6 +752,13 @@ struct alignas(64) PA_COMPONENT_EXPORT(PARTITION_ALLOC) PartitionRoot {
         internal::DirectMapAllocationGranularity());
   }
 
+  PA_ALWAYS_INLINE bool IsDirectMapped(
+      partition_alloc::internal::SlotSpanMetadata<
+          partition_alloc::internal::MetadataKind::kReadOnly>* slot_span)
+      const {
+    return IsDirectMappedBucket(slot_span->bucket);
+  }
+
   PA_ALWAYS_INLINE size_t AdjustSize0IfNeeded(size_t size) const {
     // There are known cases where allowing size 0 would lead to problems:
     // 1. If extras are present only before allocation (e.g. in-slot metadata),