CacheStorage: Fix flaky crashes with DCHECK enabled.

The cache-add.https.html WPT tests became flaky after landing
crrev.com/c/2429308.  The flakes were due to crashes accessing a vector
element beyond its length.  I believe this is due to:

  DCHECK(!blob_list_[index]);

Which is performed before we check to see if we are already stopped.
Since we clear this vector when stopped this DCHECK will crash if a
completion comes in right after aborting.

This CL moves the DCHECKs below our stopped check.  It also explicitly
sets the stopped state when successfully complete as well.

Bug: 1134021,1130781
Change-Id: I075f2daca252ac78a93d5e3991f44389adcf9689
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2443069Reviewed-by: Marijn Kruisselbrink <mek@chromium.org>
Commit-Queue: Ben Kelly <wanderview@chromium.org>
Cr-Commit-Position: refs/heads/master@{#812766}

third_party/blink/renderer/modules/cache_storage/cache.cc

@@ -205,13 +205,13 @@ class Cache::BarrierCallbackForPutResponse final
   void CompletedResponse(int index,
                          Response* response,
                          scoped_refptr<BlobDataHandle> blob) {
+    if (stopped_)
+      return;
+
     DCHECK(!response_list_[index]);
     DCHECK(!blob_list_[index]);
     DCHECK_LT(num_complete_, request_list_.size());
 
-    if (stopped_)
-      return;
-
     response_list_[index] = response;
     blob_list_[index] = std::move(blob);
     num_complete_ += 1;
@@ -223,6 +223,7 @@ class Cache::BarrierCallbackForPutResponse final
       cache_->PutImpl(resolver_, method_name_, request_list_, response_list_,
                       blob_list_, exception_state, trace_id_);
       blob_list_.clear();
+      stopped_ = true;
     }
   }