feat(c++,pt-expt): add DeepPotModelDevi C++ tests for .pt2 backend (#5342)

## Summary
- Add C++ `DeepPotModelDevi` unit tests for the pt_expt (.pt2) backend
- Generate two SE(A)+fparam/aparam .pt2 models with different seeds
(`gen_model_devi.py`) for meaningful deviations
- Both models include `default_fparam=[0.25852028]` for attribute
detection tests

## Tests added (14 total: 2 test classes × {double, float})

**TestInferDeepPotModeDeviPtExpt** (6 tests per type):
- `attrs` — cutoff, numb_types, dim_fparam==1, dim_aparam==1,
has_default_fparam
- `cpu_build_nlist` / `cpu_build_nlist_atomic` — individual dp0/dp1 vs
model_devi consistency
- `cpu_lmp_list` / `cpu_lmp_list_atomic` — same with LAMMPS nlist +
fold_back
- `cpu_lmp_list_std` — manual vs API compute_avg, compute_std_e,
compute_std_f, compute_relative_std_f

**TestInferDeepPotModeDeviPtExptPrecomputed** (1 test per type):
- `cpu_lmp_list_std` — max/min/avg std_f and max/min/mystd std_v against
hardcoded reference values

## Known limitations
- `default_fparam` **compute** path is not tested:
`DeepPotPTExpt::compute` creates a zero-size tensor when fparam is
empty, which crashes the .pt2 runner. The C++ API supports
`has_default_fparam()` detection but not automatic value substitution.

## Test plan
- [x] `gen_model_devi.py` generates two .pt2 models without errors
- [x] All 14 TYPED_TEST cases pass for both double and float
- [x] Precomputed deviation values match
- [x] No regressions in existing 94 PtExpt tests (108 total pass)
- [x] No compiler warnings
- [x] Pre-commit hooks pass

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

* **Tests**
* Added a comprehensive test suite validating multi-model deviation
inference and statistical outputs for the PyTorch-enabled backend.
* Added an automated artifact-generation step to produce paired model
artifacts used for deviation comparisons.
* Updated test setup to run the new generation step and conditionally
enable these tests only when PyTorch support is available.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: Han Wang <wang_han@iapcm.ac.cn>

Author: wanghan-iapcm