fix(pt): use eval mode in the C++ interface (#4402)

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

- **New Features**
	- Enhanced error handling for improved clarity during execution.
- Added evaluation mode setting for models to ensure correct inference
behavior.
- **Improvements**
- Streamlined device management in computation methods for better
resource utilization.
	- Improved flexibility in handling tensor creation and data transfer.
- Updated computation methods to support new logic for message passing
and mapping tensors.
- Introduced new methods for mixed-type computations, enhancing
functionality.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: Jinzhe Zeng <jinzhe.zeng@rutgers.edu>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: njzjz

source/api_cc/src/DeepPotPT.cc

@@ -88,6 +88,7 @@ void DeepPotPT::init(const std::string& model,
   }
   std::unordered_map<std::string, std::string> metadata = {{"type", ""}};
   module = torch::jit::load(model, device, metadata);
+  module.eval();
   do_message_passing = module.run_method("has_message_passing").toBool();
   torch::jit::FusionStrategy strategy;
   strategy = {{torch::jit::FusionBehavior::DYNAMIC, 10}};

source/api_cc/src/DeepSpinPT.cc

@@ -88,6 +88,7 @@ void DeepSpinPT::init(const std::string& model,
   }
   std::unordered_map<std::string, std::string> metadata = {{"type", ""}};
   module = torch::jit::load(model, device, metadata);
+  module.eval();
   do_message_passing = module.run_method("has_message_passing").toBool();
   torch::jit::FusionStrategy strategy;
   strategy = {{torch::jit::FusionBehavior::DYNAMIC, 10}};