Missing reference surfaces leak from CKF with Acts 34

[veprbl]

This is seen in detector_benchmark for acts 34 container (https://eicweb.phy.anl.gov/EIC/benchmarks/detector_benchmarks/-/jobs/4149226#L5250):

       `- JSignalHandler::handle_sigsegv(int, siginfo_t*, void*) (0x7fb59af1ca49)
        `- /usr/lib/x86_64-linux-gnu/libc.so.6 (0x7fb598c87050)
         `- Acts::Surface::getSharedPtr() const (0x7fb596161c41)
          `- eicrecon::AmbiguitySolver::process(std::vector<Acts::TrackContainer<Acts::ConstVectorTrackContainer, Acts::ConstVectorMultiTrajectory, std::shared_ptr> const*, std::allocator<Acts::TrackContainer<Acts::ConstVectorTrackContainer, Acts::ConstVectorMultiTrajectory, std::shared_ptr> const*> >, edm4eic::Measurement2DCollection const&) (0x7fb596622827)
           `- eicrecon::AmbiguitySolver_factory::Process(long, unsigned long) (0x7fb594222d3a)

Not able to reproduce locally, and even on CI it's a rare event. Looking at Ambiguity solver, it doesn't even consume trajectories, so it makes sense that effect of #1663 may miss it. But then it's not clear how it is not always failing when unpropagated track is encountered. Another question is how tracking benchmark covering $3^\circ$ to $177^\circ$ hits an issue that, to our best knowledge, is B0-specific.