Several cleanups suggested by mab.

This change addresses several issues noted by mab:
   • Use rough_time_t as a type for Roughtime timestamps. (Note that not
     all timestamps use this type. Times from the monotonic clock, for
     example, are not Roughtime timestamps.)
   • Split the Success test into so that misalignment is tested
     separately.
   • Use sizeof(tag_t) in protocol.cc.

Change-Id: I0228d35c9116aa44b9a792df54a2df6b7d26d6e3
diff --git a/BUILD b/BUILD
index 4e03973..1d3c2e1 100644
--- a/BUILD
+++ b/BUILD
@@ -44,6 +44,7 @@
 cc_library(
     name = "time_source",
     hdrs = ["time_source.h"],
+    deps = [":protocol"],
 )
 
 cc_library(