T265 deadlock when stopping the pipeline

[ankyur]

Required Info
Camera Model	T265
Firmware Version	0.2.0.951
Operating System & Version	Win10 Build 2004
Platform	PC
SDK Version	v2.38.1 (master)
Language	C++

Issue Description

Sometimes stopping the pipeline causes deadlock due to race condition.

A race condition occurs when you stopping pipeline and call tm2_sensor::stop() --> tm2_sensor::stop_interrupt() and trying to cancel last request but sometimes there are no pended request at that moment. At the same time another thread (which is processing lambda-function _interrupt_callback declared in tm2_sensor::start_interrupt()) just ready to add new request but it will never be canceled.

This code should be fixed.

    void tm2_sensor::stop_interrupt()
    {
        if (_interrupt_request) {
            if (_device->cancel_request(_interrupt_request)) {
                _interrupt_callback->cancel();
                _interrupt_request.reset();
            }
        }
    }

The simplest way to fix it is to move cancelling callback (which has own mutex) outside the cancelling request. This will cancel callback function first and will be guarantee that processing thread will be stopped. For example like this:

    void tm2_sensor::stop_interrupt()
    {
        if (_interrupt_request) {
            _interrupt_callback->cancel();
            if (_device->cancel_request(_interrupt_request))
                _interrupt_request.reset();
        }
    }

Of course if we want exactly the same behavior (actually the same log-print on stopping), we should add mutex (condition_variable) to this code.

P.S.: all the same with tm2_sensor::stop_stream() and stream request/callback.