Added host monitor thread to disconnect offline PoE devices

[themarpe]

Creates a watchdog triggered disconnect from PoE device which have gone offline.
Also addresses: #476

[themarpe]

Details are layed out in #476 and in referenced: #366 (comment)

Scenario happened as:

Stack trace

depthai.cpython-38-x86_64-linux-gnu.so!__gnu_cxx::__atomic_add(volatile _Atomic_word * __mem, int __val) (/usr/include/c++/10/ext/atomicity.h:55)
depthai.cpython-38-x86_64-linux-gnu.so!__gnu_cxx::__atomic_add_dispatch(_Atomic_word * __mem, int __val) (/usr/include/c++/10/ext/atomicity.h:96)
depthai.cpython-38-x86_64-linux-gnu.so!std::_Sp_counted_base<(__gnu_cxx::_Lock_policy)2>::_M_add_ref_copy(std::_Sp_counted_base<__gnu_cxx::_S_atomic> * this) (/usr/include/c++/10/bits/shared_ptr_base.h:142)
depthai.cpython-38-x86_64-linux-gnu.so!std::__shared_count<(__gnu_cxx::_Lock_policy)2>::__shared_count(std::__shared_count<__gnu_cxx::_S_atomic> * this, const std::__shared_count<__gnu_cxx::_S_atomic> & __r) (/usr/include/c++/10/bits/shared_ptr_base.h:740)
depthai.cpython-38-x86_64-linux-gnu.so!std::__shared_ptr<dai::ADatatype, (__gnu_cxx::_Lock_policy)2>::__shared_ptr<dai::ADatatype>(std::__shared_ptr<dai::ADatatype, __gnu_cxx::_S_atomic> * this, const std::__shared_ptr<dai::ADatatype, __gnu_cxx::_S_atomic> & __r, std::__shared_ptr<dai::ADatatype, __gnu_cxx::_S_atomic>::element_type * __p) (/usr/include/c++/10/bits/shared_ptr_base.h:1168)
depthai.cpython-38-x86_64-linux-gnu.so!std::shared_ptr<dai::ADatatype>::shared_ptr<dai::ADatatype>(std::shared_ptr<dai::ADatatype> * this, const std::shared_ptr<dai::ADatatype> & __r, std::shared_ptr<dai::ADatatype>::element_type * __p) (/usr/include/c++/10/bits/shared_ptr.h:257)
depthai.cpython-38-x86_64-linux-gnu.so!std::dynamic_pointer_cast<dai::ADatatype, dai::ADatatype>(const std::shared_ptr<dai::ADatatype> & __r) (/usr/include/c++/10/bits/shared_ptr.h:603)
depthai.cpython-38-x86_64-linux-gnu.so!dai::DataOutputQueue::get<dai::ADatatype, long, std::ratio<1l, 1000l> >(dai::DataOutputQueue * this, std::chrono::duration<long, std::ratio<1, 1000> > timeout, bool & hasTimedout) (depthai-python/depthai-core/include/depthai/device/DataQueue.hpp:227)
depthai.cpython-38-x86_64-linux-gnu.so!dai::DataOutputQueue::get<long, std::ratio<1l, 1000l> >(dai::DataOutputQueue * this, std::chrono::duration<long, std::ratio<1, 1000> > timeout, bool & hasTimedout) (depthai-python/depthai-core/include/depthai/device/DataQueue.hpp:239)
depthai.cpython-38-x86_64-linux-gnu.so!DataQueueBindings::bind(pybind11::module_&, void*)::$_2::operator()(dai::DataOutputQueue&) const(const class {...} * this, dai::DataOutputQueue & obj) (depthai-python/src/DataQueueBindings.cpp:97)
depthai.cpython-38-x86_64-linux-gnu.so!pybind11::detail::argument_loader<dai::DataOutputQueue&>::call_impl<std::shared_ptr<dai::ADatatype>, DataQueueBindings::bind(pybind11::module_&, void*)::$_2&, 0ul, pybind11::detail::void_type>(DataQueueBindings::bind(pybind11::module_&, void*)::$_2&, std::integer_sequence<unsigned long, 0ul>, pybind11::detail::void_type&&) &&(pybind11::detail::argument_loader<dai::DataOutputQueue&> * this, class {...} & f) (/home/themarpe/.hunter/_Base/cb0ea1f/84bd56f/1c0de32/Install/include/pybind11/cast.h:1441)
depthai.cpython-38-x86_64-linux-gnu.so!pybind11::detail::argument_loader<dai::DataOutputQueue&>::call<std::shared_ptr<dai::ADatatype>, pybind11::detail::void_type, DataQueueBindings::bind(pybind11::module_&, void*)::$_2&>(pybind11::detail::argument_loader<dai::DataOutputQueue&> * this, class {...} & f) (/home/themarpe/.hunter/_Base/cb0ea1f/84bd56f/1c0de32/Install/include/pybind11/cast.h:1409)
depthai.cpython-38-x86_64-linux-gnu.so!pybind11::cpp_function::initialize<DataQueueBindings::bind(pybind11::module_&, void*)::$_2, std::shared_ptr<dai::ADatatype>, dai::DataOutputQueue&, pybind11::name, pybind11::is_method, pybind11::sibling, char const*>(DataQueueBindings::bind(pybind11::module_&, void*)::$_2&&, std::shared_ptr<dai::ADatatype> (*)(dai::DataOutputQueue&), pybind11::name const&, pybind11::is_method const&, pybind11::sibling const&, char const* const&)::{lambda(pybind11::detail::function_call&)#1}::operator()(pybind11::detail::function_call&) const(const class {...} * this, pybind11::detail::function_call & call) (/home/themarpe/.hunter/_Base/cb0ea1f/84bd56f/1c0de32/Install/include/pybind11/pybind11.h:249)
depthai.cpython-38-x86_64-linux-gnu.so!pybind11::cpp_function::initialize<DataQueueBindings::bind(pybind11::module_&, void*)::$_2, std::shared_ptr<dai::ADatatype>, dai::DataOutputQueue&, pybind11::name, pybind11::is_method, pybind11::sibling, char const*>(DataQueueBindings::bind(pybind11::module_&, void*)::$_2&&, std::shared_ptr<dai::ADatatype> (*)(dai::DataOutputQueue&), pybind11::name const&, pybind11::is_method const&, pybind11::sibling const&, char const* const&)::{lambda(pybind11::detail::function_call&)#1}::__invoke(pybind11::detail::function_call&)(pybind11::detail::function_call & call) (/home/themarpe/.hunter/_Base/cb0ea1f/84bd56f/1c0de32/Install/include/pybind11/pybind11.h:224)
depthai.cpython-38-x86_64-linux-gnu.so!pybind11::cpp_function::dispatcher(PyObject * self, PyObject * args_in, PyObject * kwargs_in) (/home/themarpe/.hunter/_Base/cb0ea1f/84bd56f/1c0de32/Install/include/pybind11/pybind11.h:934)
PyCFunction_Call (Unknown Source:0)
_PyObject_MakeTpCall (Unknown Source:0)
[Unknown/Just-In-Time compiled code] (Unknown Source:0)
_PyEval_EvalFrameDefault (Unknown Source:0)
_PyEval_EvalCodeWithName (Unknown Source:0)

Code in question

    template <class T, typename Rep, typename Period>
    std::shared_ptr<T> get(std::chrono::duration<Rep, Period> timeout, bool& hasTimedout) {
        if(!running) throw std::runtime_error(exceptionMessage.c_str());
        std::shared_ptr<ADatatype> val = nullptr;
        if(!queue.tryWaitAndPop(val, timeout)) {
            hasTimedout = true;
            return nullptr;
        }
        hasTimedout = false;
        return std::dynamic_pointer_cast<T>(val); # <- Segfault - val is uninitialized (garbage value)
    }

Adding proper destructed checks and guarding those under same mutex as the rest of the LockingQueue share, should create the correct flow for cases where a queue is destructed in between messages being actively pushed into it.