fix handle spurious wakeup wrongly returning Flush too early

[hendrikmuhs]

add an atomic bool to check if flush has actually happened and condition variable did not spuriously wakeup.

fixes #339

2 important changes that fix this behavior hard-to-debug race condition:

• the condition can wake up despite not being notified, that's why we need an extra atomic bool
• the mutex should still be locked when calling notify, the problem is lifetime:
  • if the mutex is unlocked, the condition variable might get destructed after the destruction of the mutex, by moving it in scope, the lambda is fully destructed before the consumer side kicks in. The consumer holds the mutex that's why the mutex is in scope when the condition variable goes out of scope
  • error: std::__1::system_error: mutex lock failed: Invalid argument

Note:

• I've run many loops here: c1af918 to test this, reverted for this PR
• I think this race condition happened only for clang / on mac, I haven't checked if clang on Linux would show the same thing, likely yes