Sync from internal Google development.

Many improvements, too many to mention.  One significant
perf regression warrants investigation:

omitfp.parsetoproto2_googlemessage1.upb_jit: 343 -> 252 (-26.53)
plain.parsetoproto2_googlemessage1.upb_jit: 334 -> 251 (-24.85)

25% regression for this benchmark is bad, but since I don't think
there's any fundamental design issue that caused it I'm going to
go ahead with the commit anyway.  Can investigate and fix later.

Other benchmarks were neutral or showed slight improvement.

1 files changed, 15 insertions, 9 deletions

diff --git a/tests/test.proto b/tests/test.proto
index f3dde24..e634ed2 100644
--- a/tests/test.proto
+++ b/tests/test.proto
@@ -1,14 +1,10 @@
 
 // A series of messages with various kinds of cycles in them.
-//      +-+---+    +---+
-//      V |   |    |   |
-// A -> B-+-> C -> D<--+
-// ^          |    |
-// +----------+----+
-//
-// This tests the following cases:
-//  - B and C are together in multiple cycles
-//  - B and D are cycles to themselves.
+//      +-+---+    +---+    +---+
+//      V |   |    V   |    V   |
+// A -> B-+-> C -> D---+--->E---+
+// ^          |`---|--------^
+// +----------+----+        F
 
 message A {
   optional B b = 1;
@@ -23,11 +19,21 @@ message C {
   optional A a = 1;
   optional B b = 2;
   optional D d = 3;
+  optional E e = 4;
 }
 
 message D {
   optional A a = 1;
   optional D d = 2;
+  optional E e = 3;
+}
+
+message E {
+  optional E e = 1;
+}
+
+message F {
+  optional E e = 1;
 }
 
 // A proto with a bunch of simple primitives.