refactor: Use pointer for (C)KF inputMeasurements

[XiaocongAi]

This PR changes the inputMeasurements in (C)KF actor to use pointer. This could save the data copy time vastly.

[XiaocongAi]

A comparison of the timing test for (C)KF using smeared truth parameters for ttbar events W/O the change below:

[codecov]

Codecov Report

Merging #742 (734c632) into master (a9ec279) will not change coverage.
The diff coverage is 72.72%.

@@           Coverage Diff           @@
##           master     #742   +/-   ##
=======================================
  Coverage   48.86%   48.86%           
=======================================
  Files         325      325           
  Lines       16682    16682           
  Branches     7791     7791           
=======================================
  Hits         8151     8151           
  Misses       3061     3061           
  Partials     5470     5470           

Impacted Files	Coverage Δ
...de/Acts/TrackFinding/CombinatorialKalmanFilter.hpp	29.21% <66.66%> (ø)
Core/include/Acts/TrackFitting/KalmanFitter.hpp	40.00% <75.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a9ec279...734c632. Read the comment docs.

[gagnonlg]

Hello @XiaocongAi , just a suggestion: std::unique_ptr would probably be safer here. https://en.cppreference.com/w/cpp/memory/unique_ptr

[asalzburger]

Wow, looks like we spent a lot of time copying into containers, right?

As discussed: wherever we hit a performance wall, we should use bare pointers, etc.

Can you please specify though that the ownership of the container is NOT with the KalmanActor?

[asalzburger]

Hello @XiaocongAi , just a suggestion: std::unique_ptr would probably be safer here. https://en.cppreference.com/w/cpp/memory/unique_ptr

Hi, I have had a similar comment bout, however, I am not sure a std::unique_ptr would work, as you get the inputMeasurements object and then just dereference it?

Let's try to do the following:

• if there's really only one unique measurement container per KF/CKF run, then yes, we should try to make a unique_ptr.
  @XiaocongAi - can you clarify this, please?

[asalzburger]

We should try to clarify the ownership of the container.

[XiaocongAi]

Hello @XiaocongAi , just a suggestion: std::unique_ptr would probably be safer here. https://en.cppreference.com/w/cpp/memory/unique_ptr

Hi, I have had a similar comment bout, however, I am not sure a std::unique_ptr would work, as you get the inputMeasurements object and then just dereference it?

Let's try to do the following:

• if there's really only one unique measurement container per KF/CKF run, then yes, we should try to make a unique_ptr.
  @XiaocongAi - can you clarify this, please?

Yes, the inputMeasurements is created per KF/CKF invocation. The KF/CKF fit/findTracks function has ownership of the inputMeasurements. The Kalman actor just deference it without any ownership of it. Since we know that the Kalman actor is created inside the fit/findTracks, i.e. the lifetime of the actor won't be longer than the inputMeasurements, I guess we should be safe with a bare pointer.

[asalzburger]

Merged 'master' in, when it's done, I will merge it in.