Accelerating Linear Recurrent Neural Networks for the Edge with Unstructured Sparsity