[BUG] model_devi_f_avg_relative crashes when model_devi_avgf.out is matched as a PIMD bead output

[SchrodingersCattt]

Bug summary

With model_devi_f_avg_relative=True, DP-GEN writes an additional model_devi_avgf.out file into every model-deviation task directory. On the next selection step, _read_model_devi_file() currently uses glob("model_devi*.out") and assumes every match has the form model_devi<digits>.out. The generated normalization file does not match that pattern, so selection crashes with:

AttributeError: 'NoneType' object has no attribute 'group'

Absolute model deviation does not trigger this because only model_devi.out exists.

Environment

• DP-GEN 0.13.4.dev17+gd481bfb8b
• DeePMD-kit 3.2.0
• PyTorch-exportable PT2 models
• model_devi_f_avg_relative=True
• model_devi_f_trust_lo=0.5, model_devi_f_trust_hi=1.0

Reproduction

1. Run a LAMMPS model-deviation task with model_devi_f_avg_relative=True and model_devi_merge_traj=True.
2. Let DP-GEN write both model_devi.out and model_devi_avgf.out.
3. Resume or reach make_fp, which calls _select_by_model_devi_standard().
4. _read_model_devi_file() matches both files and the numeric-suffix regex fails.

Suggested fix

Match the base file separately and only match numbered bead outputs, e.g. model_devi.out plus model_devi[0-9]*.out, or make the sort key accept the base filename. Add a regression test where model_devi.out and model_devi_avgf.out coexist.

I reproduced this in a real Neo DP-GEN run and applied the minimal file-match fix locally; the relative selection then completed successfully.