tp: record perf_sample source stops due to a guardrail as a stats entry

Simply put:
  perf_guardrail_stop[perf_session_id]: timestamp

Using perf_session_id (as already used by perf_sample and
perf_counter_track tables) as an index for the possible case of multiple
data sources in a trace.

Change-Id: I1e7c751ac666a02dcc296cc7433f091cf616b34a

src/trace_processor/storage/stats.h

diff --git a/src/trace_processor/storage/stats.h b/src/trace_processor/storage/stats.h
index 16d6e63..28869c5 100644
--- a/src/trace_processor/storage/stats.h
+++ b/src/trace_processor/storage/stats.h
@@ -176,7 +176,8 @@
       "An ftrace packet was seen before the tracing start timestamp from "     \
       "the tracing service. This happens if the ftrace buffers were not "      \
       "cleared properly. These packets are silently dropped by trace "         \
-      "processor.")
+      "processor."),                                                           \
+  F(perf_guardrail_stop_ts,             kIndexed, kDataLoss, kTrace,    "")
 // clang-format on
 
 enum Type {