Exploring the Influence of Dimensionality Reduction on Anomaly Detection Performance in Multivariate Time Series