[Code scan] RDF helpers mutate sel_type and poison later default calls

This issue is part of a Codex global repository code scan. The RDF functions use mutable `sel_type=[None, None]` defaults and `_compute_rdf_1frame()` rewrites the list in place. A default call can therefore change `compute_rdf.__defaults__`, causing later default calls on a different system to reuse stale atom types and return invalid values. Affected code: https://github.com/deepmodeling/dpdata/blob/a7a50bfc096e1a65470af6f1c48152dc0275ec1c/dpdata/md/rdf.py#L43-L50 https://github.com/deepmodeling/dpdata/blob/a7a50bfc096e1a65470af6f1c48152dc0275ec1c/dpdata/md/rdf.py#L61-L70 Minimal reproducer: ```python import numpy as np from dpdata.md import rdf as rdfmod box = np.eye(3).reshape(1, 3, 3) * 10 pos = np.array([[[0.0, 0.0, 0.0], [1.0, 0.0, 0.0]]]) rdfmod.compute_rdf(box, pos, np.array([0, 1])) print(rdfmod.compute_rdf.__defaults__) _, stat, _ = rdfmod.compute_rdf(box, pos, np.array([2, 2])) print(np.isfinite(stat).all(), stat[:3]) ``` Current behavior: ```text ([[np.int64(0), np.int64(1)], [np.int64(0), np.int64(1)]], 5, 100) False [nan nan nan] ``` The function should not mutate either its default object or a caller-provided `sel_type` list. A common fix is to default to `None` and construct a local normalized selection list per call/frame.