feat(pt): pass mapping from LAMMPS to PT (#4351)

As discussed, this PR passes mapping from LAMMPS to the PT C++
interface, which is helpful for the external GNN models.

The mapping interface is synced from #4307.

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

## Release Notes

- **New Features**
- Introduced a new function `DP_NlistSetMapping` for setting mappings in
neighbor lists.
- Added `set_mapping` method to `InputNlist` for mapping atoms to real
atoms.
- Enhanced `compute` methods in `DeepPotPT`, `PairDeepMD`, and `FixDPLR`
classes to support new mapping functionalities.

- **Bug Fixes**
- Improved error handling in various classes to ensure robustness during
execution.

- **Documentation**
- Updated and added comments for clarity and consistency in new and
existing functions.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: Jinzhe Zeng <[email protected]>

source/api_cc/src/DeepPotPT.cc

@@ -203,6 +203,15 @@ void DeepPotPT::compute(ENERGYVTYPE& ener,
       comm_dict.insert("recv_num", recvnum_tensor);
       comm_dict.insert("communicator", communicator_tensor);
     }
+    if (lmp_list.mapping) {
+      std::vector<std::int64_t> mapping(nall_real);
+      for (size_t ii = 0; ii < nall_real; ii++) {
+        mapping[ii] = lmp_list.mapping[fwd_map[ii]];
+      }
+      mapping_tensor =
+          torch::from_blob(mapping.data(), {1, nall_real}, int_option)
+              .to(device);
+    }
   }
   at::Tensor firstneigh = createNlistTensor(nlist_data.jlist);
   firstneigh_tensor = firstneigh.to(torch::kInt64).to(device);