proto: consistently use non-nil, zero-length []bytes for empty bytes strings

The fast-path decoder decodes zero-length repeated bytes values as
non-nil, zero-length []bytes. Do the same in the reflection decoder.

This isn't really a correctness issue, since there's no ambiguity about what a
nil entry in a [][]byte means. Still a good idea for consistency, and
retains v1 behavior.

Change-Id: Icd2cb726d14ff1f2b9f142e65756777a359971f3
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/210257
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
diff --git a/proto/decode_gen.go b/proto/decode_gen.go
index 61d5f8a..c40c01d 100644
--- a/proto/decode_gen.go
+++ b/proto/decode_gen.go
@@ -167,7 +167,7 @@
 		if n < 0 {
 			return val, 0, wire.ParseError(n)
 		}
-		return protoreflect.ValueOfBytes(append(([]byte)(nil), v...)), n, nil
+		return protoreflect.ValueOfBytes(append(emptyBuf[:], v...)), n, nil
 	case protoreflect.MessageKind:
 		if wtyp != wire.BytesType {
 			return val, 0, errUnknown
@@ -564,7 +564,7 @@
 		if n < 0 {
 			return 0, wire.ParseError(n)
 		}
-		list.Append(protoreflect.ValueOfBytes(append(([]byte)(nil), v...)))
+		list.Append(protoreflect.ValueOfBytes(append(emptyBuf[:], v...)))
 		return n, nil
 	case protoreflect.MessageKind:
 		if wtyp != wire.BytesType {
@@ -598,3 +598,6 @@
 		return 0, errUnknown
 	}
 }
+
+// We append to an empty array rather than a nil []byte to get non-nil zero-length byte slices.
+var emptyBuf [0]byte