Fix a string lifetime bug in perfetto support code

This is a speculative fix for the test failures in
https://crbug.com/923166#c40 based on the stack in the logcat and code
inspection. AndroidSystemProducer stores a raw const char* pointer
passed in from many layers up the stack.

In particular, the SetSystemProducerSocketAndChecksAsync function used
in tests passes in producer_string.c_str() where producer_string is a
std::string that was bound into a callback somewhere. This is a UAF and
I suspect it is what's tripping the libc++ debugging assertion.

Bug: 923166
Change-Id: Id9153009c34ef942a2d9a4b21c760c0394258f54
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1924235
Auto-Submit: David Benjamin <davidben@chromium.org>
Reviewed-by: Stephen Nusko <nuskos@chromium.org>
Reviewed-by: Eric Seckler <eseckler@chromium.org>
Commit-Queue: Eric Seckler <eseckler@chromium.org>
Cr-Commit-Position: refs/heads/master@{#719870}

services/tracing/public/cpp/perfetto/android_system_producer.cc

@@ -405,7 +405,7 @@ void AndroidSystemProducer::ActivateTriggers(
 void AndroidSystemProducer::ConnectSocket() {
   state_ = State::kConnecting;
   service_ = perfetto::ProducerIPCClient::Connect(
-      socket_name_, this,
+      socket_name_.c_str(), this,
       base::StrCat(
           {mojom::kPerfettoProducerNamePrefix,
            base::NumberToString(

services/tracing/public/cpp/perfetto/android_system_producer.h

@@ -137,7 +137,7 @@ class COMPONENT_EXPORT(TRACING_CPP) AndroidSystemProducer
   // later.
   void DelayedReconnect();
 
-  const char* socket_name_;
+  std::string socket_name_;
   uint32_t connection_backoff_ms_;
   uint64_t data_sources_tracing_ = 0;
   bool disallow_pre_android_pie = true;