Batched QR and SVD Algorithms on GPUs with Applications in Hierarchical Matrix Compression