A shrinkage principle for heavy-tailed data: High-dimensional robust low-rank matrix recovery | Synapse