[stubs] Micro optimizations to CodeAssember's allocation path

Now that the GC team has landed the appropriate changes to ensure that the top
page of the address space is never used for allocation, the inlined fast-case
allocation path in the CodeAssembler can be micro-optimized to an add to top
followed by an unsigned compare to limit, eliding a no-longer-needed overflow
check.

Review-Url: https://codereview.chromium.org/1923803003
Cr-Commit-Position: refs/heads/master@{#35830}

src/code-stub-assembler.cc

@@ -322,7 +322,8 @@ Node* CodeStubAssembler::AllocateRawUnaligned(Node* size_in_bytes,
   Label runtime_call(this, Label::kDeferred), no_runtime_call(this);
   Label merge_runtime(this, &result);
 
-  Branch(IntPtrLessThan(IntPtrSub(limit, top), size_in_bytes), &runtime_call,
+  Node* new_top = IntPtrAdd(top, size_in_bytes);
+  Branch(UintPtrGreaterThanOrEqual(new_top, limit), &runtime_call,
          &no_runtime_call);
 
   Bind(&runtime_call);
@@ -342,7 +343,7 @@ Node* CodeStubAssembler::AllocateRawUnaligned(Node* size_in_bytes,
   Bind(&no_runtime_call);
   Node* no_runtime_result = top;
   StoreNoWriteBarrier(MachineType::PointerRepresentation(), top_address,
-                      IntPtrAdd(top, size_in_bytes));
+                      new_top);
   no_runtime_result = BitcastWordToTagged(
       IntPtrAdd(no_runtime_result, IntPtrConstant(kHeapObjectTag)));
   result.Bind(no_runtime_result);