Further refactoring and cleanup of TrkGlobalChi2Fitter

In this merge request, we do some further refactoring and cleaning up of the TrkGlobalChi2Fitter class. In short, we make the following changes:

1. We migrate several instances of std::vector<double> and even std::vector<std::vector<double>> to Amg::VectorX and Amg::MatrixX, respectively. The point of that is to improve readability, performance, and interoperability with Eigen classes and methods.
2. Replacement of some naively implemented matrix-matrix methods to use Eigen, which is possible now do the previous point.
3. Deduplication of the calculateDerivative method, as well as the track making code, to reduce its code volume and improve clarity.
4. Removal of some mysterious and unused method declarations.

I've done my best to validate this change, and the results seem to be unchanged on a q431 test.

WARNING: big files (>100K) are found in the changeset

284K in file Tracking/TrkFitter/TrkGlobalChi2Fitter/src/GlobalChi2Fitter.cxx

This merge request affects 1 package:

• Tracking/TrkFitter/TrkGlobalChi2Fitter

Adding @amorley as watcher

added Tracking master review-pending-level-1 labels

CI Result SUCCESS (hash 4e0db628)

	Athena	AthSimulation	AthGeneration	AnalysisBase
externals
cmake
make
required tests
optional tests

Full details available on this CI monitor view
Athena: number of compilation errors 0, warnings 0
AthSimulation: number of compilation errors 0, warnings 0
AthGeneration: number of compilation errors 0, warnings 0
AnalysisBase: number of compilation errors 0, warnings 0
For experts only: Jenkins output [CI-MERGE-REQUEST-CC7 17861]

From what I can tell the CMake warning is not related to this MR.

indeed this is another instance of ATLINFR-3652; I shall approve this as a sensible translation of the original code, but would like to be sure also that the timing has not suffered as a result.

added review-approved label and removed review-pending-level-1 label

To guard against performance regressions I ran q431 with this version of the code as well as ac6c952e (tag nightly/master/2020-07-27T0451 from which this branches off) and the results are that PerfMon reports a runtime of 923 milliseconds in the InDetSiSpTrackFinder for the old version and 913 milliseconds for this merge request. That's a speed-up of 1%, probably just a product of timing fluctuation.

merged

mentioned in commit 4ccfcc8f

added sweep:ignore label

mentioned in merge request !35247 (merged)

resolved all threads