Fix bug with creating PBPMap from multi-layer TensorMap

[jadball]

Unlikely to have caused problems in current usage - regular notebook workflow doesn't initiate a refinement from a TensorMap with more than one z layer (z stacking comes later). In the previous code, mi was always 0, because recon_index_to_map (correctly) defaults to a 2D slice. We should be checking the self.phase_ids phase mask at the user-supplied z_layer value rather than 0.