Don't unlock when locking failed in the timer thread

Fixes an issue found by helgrind:

29171== Thread #2 unlocked lock at 0x242A60 currently held by thread #1

29171== at 0x4C30156: mutex_unlock_WRK (hg_intercepts.c:1086)

29171== by 0x4C33ACC: pthread_mutex_unlock (hg_intercepts.c:1107)

29171== by 0x2005B9: __gthread_mutex_unlock(pthread_mutex_t*) (gthr-default.h:778)

29171== by 0x203493: std::mutex::unlock() (std_mutex.h:121)

29171== by 0x200C78: (anonymous namespace)::HeapTrack::~HeapTrack() (libheaptrack.cpp:309)

29171== by 0x2016C8: (anonymous namespace)::HeapTrack::LockedData::LockedData(_IO_FILE*, void (*)())::{lambda()#1}::operator()() const (libheaptrack.cpp:631)

29171== by 0x202DD0: void std::invoke_impl<void, (anonymous namespace)::HeapTrack::LockedData::LockedData(_IO_FILE*, void (*)())::{lambda()#1}>(std::invoke_other, (anonymous namespace)::HeapTrack::LockedData::LockedData(_IO_FILE*, void (*)())::{lambda()#1}&&) (invoke.h:60)

Use an exception to get out of the constructor to prevent
the destructor from being called which would always unlock
the mutex, even if the timer thread didn't get the lock itself.