[Code scan] Dict cutoff validation compares whole dataset dicts to scalar model cutoffs

[njzjz]

This issue is part of a Codex global repository scan.

Problem:
When model cutoffs are dictionaries, check_cutoffs() verifies that each key exists in the dataset cutoff dict, but then compares the whole dataset dict to a scalar value. The same pattern appears for r_max, er_max, and oer_max.

Code references:
https://github.com/deepmodeling/deeptb/blob/86c60c73996f0dd961c3138f2e88424382cb734e/dptb/data/build.py#L334-L340
https://github.com/deepmodeling/deeptb/blob/86c60c73996f0dd961c3138f2e88424382cb734e/dptb/data/build.py#L346-L353
https://github.com/deepmodeling/deeptb/blob/86c60c73996f0dd961c3138f2e88424382cb734e/dptb/data/build.py#L362-L369

Impact:
Valid dict-mode cutoff configurations raise TypeError instead of being validated, so model/dataset compatibility checks fail for per-type or per-bond cutoff settings.

Suggested fix:
Compare self.r_max[key], self.er_max[key], and self.oer_max[key] to the corresponding model cutoff values.