mojo_unittests: Decrease cost of MessageTest.ExtendMessagePayloadLarge

This test is timing out on Windows debug, where even on very fast
devices it can take 35+ seconds to run.

The runtime grows linearly with the size of the test message used, so
this cuts that size to one tenth (512 kB instead of 5 MB).

TBR=jcivelli@chromium.org

Bug: 819046
Change-Id: Ib4598981856566dc8ae54de1060840b52e30b959
Reviewed-on: https://chromium-review.googlesource.com/954045Reviewed-by: Ken Rockot <rockot@chromium.org>
Commit-Queue: Ken Rockot <rockot@chromium.org>
Cr-Commit-Position: refs/heads/master@{#541704}

mojo/edk/system/message_unittest.cc

@@ -705,11 +705,11 @@ TEST_F(MessageTest, ExtendMessagePayloadLarge) {
     ASSERT_GE(buffer_size, static_cast<uint32_t>(kTestMessageHeader.size()));
     memcpy(buffer, kTestMessageHeader.data(), kTestMessageHeader.size());
 
-    // 5 MB should be well beyond any reasonable default buffer size for the
+    // 512 kB should be well beyond any reasonable default buffer size for the
     // system implementation to choose, meaning that this test should guarantee
     // several reallocations of the serialized message buffer as we
     // progressively extend the payload to this size.
-    constexpr size_t kTestMessagePayloadSize = 5 * 1024 * 1024;
+    constexpr size_t kTestMessagePayloadSize = 512 * 1024;
     std::vector<uint8_t> test_payload(kTestMessagePayloadSize);
     base::RandBytes(test_payload.data(), kTestMessagePayloadSize);