Updated negative determinant error handling in KalmanVertexOnJetAxisUpdator
This MR addresses a rare issue observed in extreme cases of compression of the off-diagonal elements of the track covariance matrix (beyond the one intended to be use for Run 3), which was observed in https://its.cern.ch/jira/browse/AFT-508?focusedCommentId=3615161&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-3615161
Due to the finite numerical precision, the matrix inversion can yield a negative determinant while the original matrix has a positive determinant, which can trigger an error in the KalmanVertexOnJetAxisUpdator. The error is demoted to a warning in such cases, while an error is still reported if both the inverse and the original matrix have a negative determinant.