MsgStack and PetscLib thread safety

[ZedThree]

I've not actually managed to make a test for MsgStack that fails without this change and passes with it. For example:

TEST(MsgStackTest, ThreadSafety) {
  MsgStack msg_stack;

  const std::string expected_dump = "====== Back trace ======\n";

  BOUT_OMP(parallel for)
  for (int i = 0; i < 1000; i++) {
    int id = msg_stack.push("Message", i);
    msg_stack.pop(id);
  }
  auto dump = msg_stack.getDump();
  EXPECT_EQ(dump, expected_dump);
}

looks like something this change should fix, but doesn't.

@johnomotani Any chance you've got an example of something that was failing before?

We could use standard library std::lock_guard with a custom mutex here, that might also let us be thread safe for other types of threads, if anyone is ever likely to use them.

[johnomotani]

@johnomotani Any chance you've got an example of something that was failing before?

I don't think I ever had a test that was failing because of MsgStack, it was just that TSan was complaining about it.