Ca: bugfix with nans in the track fit

A fix for the redmine issue: https://redmine.cbm.gsi.de/issues/2981#change-14601

Edit for Redmine linking: Refs #2981

algo/ca/core/data/CaMeasurementXy.h

@@ -175,18 +175,23 @@ namespace cbm::algo::ca
     ///------------------------------
     /// Data members
 
-    DataT fX {constants::Undef<DataT>};    ///< x coordinate of the measurement
-    DataT fY {constants::Undef<DataT>};    ///< y coordinate of the measurement
-    DataT fDx2 {constants::Undef<DataT>};  ///< rms^2 of the x coordinate measurement
-    DataT fDy2 {constants::Undef<DataT>};  ///< rms^2 of the y coordinate measurement
-    DataT fDxy {constants::Undef<DataT>};  ///< covariance of the x and y coordinate measurements
+    // Initializing parameters with NANs spoils the track fit where
+    // the masked-out SIMD entries are suppressed by a multication by 0.
+    // Therefore, we initialize the data members here with finite numbers.
+    // For the numerical safety, with some reasonable numbers.
+
+    DataT fX {0.};    ///< x coordinate of the measurement
+    DataT fY {0.};    ///< y coordinate of the measurement
+    DataT fDx2 {1.};  ///< rms^2 of the x coordinate measurement
+    DataT fDy2 {1.};  ///< rms^2 of the y coordinate measurement
+    DataT fDxy {0.};  ///< covariance of the x and y coordinate measurements
 
     /// number of degrees of freedom (used for chi2 calculation)
     /// if ndf == 1, the measurement is used in the chi2 calculation
     /// if ndf == 0, the measurement is not used in the chi2 calculation
 
-    DataT fNdfX = constants::Undef<DataT>;  ///< ndf for the x coordinate measurement
-    DataT fNdfY = constants::Undef<DataT>;  ///< ndf for the y coordinate measurement
+    DataT fNdfX = 0.;  ///< ndf for the x coordinate measurement
+    DataT fNdfY = 0.;  ///< ndf for the y coordinate measurement
 
   } _fvecalignment;