[Code scan] Preserve caller-provided conf_idx in make_lmp_task_group_from_config

[njzjz]

Found by Codex global repository scan of deepmodeling/dpgen2 at commit 2679611a3704f5c2646c8cb353e34177518db758.

make_lmp_task_group_from_config overwrites caller-provided conf_idx with None:

dpgen2/dpgen2/exploration/task/make_task_group_from_config.py

Lines 631 to 640 in 2679611

	def make_lmp_task_group_from_config(
	numb_models,
	mass_map,
	config,
	):
	# Work around the required conf_idx.
	# May not be a good design!!!
	config["conf_idx"] = [] if "conf_idx" not in config else None
	config = lmp_normalize(config)
	config = config_strip_confidx(config)

The comment says this is a workaround for required conf_idx, but the conditional is reversed for preserving user input:

config["conf_idx"] = [] if "conf_idx" not in config else None

If conf_idx is present, the helper discards it before normalization and before config_strip_confidx can process it. Direct users of the exported helper can lose their selected configuration indices or fail validation with None.

Suggested fix: preserve an existing conf_idx value, only insert the empty default when the key is absent, and add a regression test that passes an explicit conf_idx through this helper.