Fix PairwiseGP test that fails on CUDA (#3283)

Summary:
Pull Request resolved: #3283

**Context**: `test_load_state_dict_recomputes_laplace` from `botorch.test.models.test_pairwise_gp.TestPairwiseGP` has been failing since it was added (D99173855) due to an issue with tensors on different devices.

**This PR**: Moves a tensor to the appropriate device

Reviewed By: ItsMrLin

Differential Revision: D101675014

fbshipit-source-id: ca91dd5611d706f9b8ddc8ff3269567e9aa62f72

Author: esantorella

botorch/models/pairwise_gp.py

@@ -887,7 +887,7 @@ def set_train_data(
         self.D = torch.zeros(
             D_size, dtype=self.datapoints.dtype, device=self.datapoints.device
         )
-        comp_view = self.comparisons.view(-1, self.m, 2).long()
+        comp_view = self.comparisons.view(-1, self.m, 2).long().to(self.D.device)
         for i, sub_D in enumerate(self.D.view(-1, self.m, self.n)):
             sub_D.scatter_(1, comp_view[i, :, [0]], 1)
             sub_D.scatter_(1, comp_view[i, :, [1]], -1)