Sparse Spiking Neural Network: Exploiting Heterogeneity in Timescales for Pruning Recurrent SNN